WO2018133851A1 - Point cloud data processing method and apparatus, and computer storage medium - Google Patents

Abstract

A point cloud data processing method and apparatus, and a computer storage medium. The method comprises: classifying frames of collected point cloud data to obtain low-ground-object point cloud data (201); extracting candidate guardrail point cloud data from the low-ground-object point cloud data in the direction of a driving track (202); spatially clustering the candidate guardrail point cloud data to obtain candidate guardrail point cloud sets (203); identifying the candidate guardrail point cloud sets obtained in the spatial clustering, so as to obtain point cloud data of a central separation guardrail of a road (204); and performing three-dimensional curve fitting on the point cloud data of the central separation guardrail of the road, so as to obtain data of the central separation guardrail of the road, which is indicated in a high-precision map (205).

Description

Point cloud data processing method, device and computer storage medium

Cross-reference to related applications

The present application is filed on the basis of the Chinese Patent Application No. PCT Application No.

Technical field

The present invention relates to electronic map technology, and in particular to a point cloud data processing method and apparatus, and a computer storage medium.

Background technique

As a typical application mode of artificial intelligence, autonomous driving technology is receiving unprecedented attention. Then, the lack of high-precision maps has become one of the bottlenecks of automatic driving in various countries. The central segment of the road with guardrail data, as an important part of the high-precision map, is an indispensable part, and is a prerequisite for achieving high-precision vehicle positioning and automatic driving safety. A typical example of a road center divider with a guardrail is shown in Figure 1.

At present, the common method for extracting the guardrail from the central divider of the road is the manual extraction method and Baidu's recently-published single-frame point cloud-based extraction method (Chinese Patent Application No. 201511025864.x, the patent name is the protection point point cloud extraction method and device). The manual extraction method requires the internal operation personnel to open the road point cloud data in the special software, and manually extract the fence point cloud data. However, the manual extraction method has the following disadvantages: the manual extraction method has large amount of point cloud data and complicated operation, resulting in large extraction workload and low efficiency; high cost and large-scale operation cannot be performed: due to low manual extraction efficiency, massive point cloud data A huge amount of manual work is required, which makes it impossible to promote the work on a large scale. The single frame point cloud based extraction method uses the preset guard point cloud feature to extract and identify the single frame point cloud data to obtain the final guard point cloud data. However, the extraction method based on single-frame point cloud has the following disadvantages: the use of single-frame point cloud for the extraction and recognition of guardrail features, due to the sparsity inherent in single-frame point cloud data, the preset guardrail features are not obvious, and it is easy to cause errors and omissions. Extraction; only considering the shape characteristics of the guardrail, does not consider the spatial topological relationship between the guardrail and other features; in addition, the method does not distinguish between the central divider of the road and the extraction of the guardrails on both sides of the road, but in the practical application of unmanned driving The application requirements of these two types of guardrails are not the same, and need to be distinguished when extracting.

Summary of the invention

In view of this, embodiments of the present invention are directed to providing a point cloud data processing method and apparatus, and a computer storage medium, which can efficiently and accurately extract point cloud data of a road center divider guardrail, thereby improving the robustness of identification and extraction.

The technical solution of the present invention is implemented as follows:

According to a first aspect of the embodiments of the present invention, a point cloud data processing method is provided. The method includes: classifying collected point cloud data of each frame to obtain low ground point cloud data; The direction extracts the candidate guardrail point cloud data from the low ground point cloud data; spatial clusters the candidate guardrail point cloud data to obtain the candidate guardrail point cloud set; and the candidate guardrail point cloud set obtained by the clustering Identifying, obtaining point cloud data of the road center divider with guardrail; performing three-dimensional curve fitting on the road center divider with guardrail point cloud data, and obtaining road central divider guardrail data represented in the high-precision map.

According to a second aspect of the embodiments of the present invention, a point cloud data processing method is provided, including the steps of:

Obtaining low-land point cloud data based on at least position information in each frame point cloud data;

Extracting candidate guardrail point cloud data from the low ground point cloud data according to a distance between the low ground object and a specific side of the vehicle along a direction of the vehicle track;

Spatial clustering of candidate guardrail point cloud data according to preset conditions, to obtain a set of candidate guardrail point cloud clouds;

At least spatial feature recognition is performed on each candidate barrier point cloud set obtained by clustering, and point cloud data of the road center partition guardrail is obtained;

Three-dimensional curve fitting is performed on the point cloud data of the central divider of the road, and the road center divider guardrail data represented in the high-precision map is obtained.

According to a third aspect of the present invention, a point cloud data processing apparatus is provided. The apparatus includes: a classification unit configured to classify the collected point cloud data to obtain low ground point cloud data. a first extracting unit configured to extract candidate guardrail point cloud data from the low ground point cloud data along a direction of the vehicle track; the clustering unit is configured to perform spatial clustering on the candidate guardrail point cloud data, Obtaining a set of candidate guardrail point clouds; the identifying unit is configured to identify each candidate guardrail point cloud set obtained by the cluster, and obtain point cloud data of the road center divider guardrail; the fitting unit is configured to be separated from the road center The 3D curve fitting is performed with the guardrail point cloud data, and the road center divider guardrail data represented in the high-precision map is obtained.

According to a fourth aspect of the present invention, a computer storage medium storing computer executable instructions for executing point cloud data according to an embodiment of the present invention is provided. Approach.

According to the technical solution of the embodiment of the present invention, the collected point cloud data of each frame is classified to obtain low point object cloud data; and the candidate is extracted from the low ground point cloud data along the direction of the vehicle track Guard point cloud data; spatial clustering of candidate guardrail point cloud data to obtain a set of candidate guardrail point cloud; identify the candidate guardrail point cloud collection obtained by clustering, and obtain point cloud data of the road center divider guardrail; The road center partitions the guardrail point cloud data for three-dimensional curve fitting, and obtains the road center separation zone guardrail data represented in the high-precision map; thus, the road central separation zone guardrail can be quickly and automatically extracted from the vehicle laser point cloud. The data, in turn, provides basic data for high-end applications such as vehicle-assisted positioning and driverless driving, which can greatly improve the automatic extraction efficiency of the road center divider with guardrails, reduce the amount of manual work, and reduce the production cost of high-precision maps.

DRAWINGS

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly described below. It is obvious that the drawings in the following description are only some embodiments of the present invention. Other drawings may also be obtained from those of ordinary skill in the art in light of the inventive work.

1 is a schematic view of a road center divider belt guardrail according to an embodiment of the present invention;

2 is a schematic flowchart of a method for processing a point cloud data according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of a typical single frame laser scan line data according to an embodiment of the present invention; FIG.

4 is a side view of a road center divider with a guardrail according to an embodiment of the present invention;

FIG. 5 is a schematic diagram showing a vertical projection of a road center divider guard rail to a two-dimensional plane according to an embodiment of the present invention; FIG.

FIG. 6 is a three-dimensional curve diagram of data of a road center divider with guard rails according to an embodiment of the present invention; FIG.

FIG. 7 is a schematic structural diagram of a point cloud data processing apparatus according to an embodiment of the present disclosure;

FIG. 8 is a schematic structural diagram of an optional software and hardware of a point cloud data processing apparatus according to an embodiment of the present invention; FIG.

9-1 is a schematic diagram of an optional scenario of point cloud data processing when the point cloud data processing device is distributed in the cloud according to an embodiment of the present invention;

9-2 is a schematic diagram of an optional scenario of point cloud data processing when the point cloud data processing device is distributed on the vehicle side according to an embodiment of the present invention;

FIG. 10 is an optional schematic flowchart of a point cloud data processing method according to an embodiment of the present invention.

detailed description

In order to understand the features and technical contents of the present invention in more detail, the important abbreviations and key terms involved in the present invention are first described below.

Point cloud data refers to data that is scanned and recorded in the form of dots by a scanning device (such as a laser scanner) installed in a vehicle or other mobile device (such as an aircraft), each point containing a three-dimensional point Coordinates, as well as attribute information of corresponding 3D points, such as red, green, blue (RGB, Red, Green, Blue) color information, or reflection intensity information (Intensity).

Vehicle laser point cloud: Point cloud data collected by a scanning device installed on a mobile measuring vehicle.

High-precision maps: maps that represent lane-level maps, including lane lines, markings, and road parameters. It has at least centimeter-level positioning accuracy and can also have road facility information (such as traffic lights, electronic eyes, and traffic signs). Among them, the road parameters can be static traffic information (such as whether the road is restricted or not), or dynamic traffic information such as traffic flow (whether it is unblocked, whether there is a traffic accident), road conditions (whether there is water, ice) Wait).

Road facilities, auxiliary facilities near the road that are continuously distributed along the road, such as road guardrails, traffic signs, traffic lights, and electronic eyes.

Guardrail: A longitudinal energy absorbing structure that absorbs collision energy by self-deformation or vehicle climb, thereby changing the direction of the vehicle, preventing the vehicle from going out of the road or entering the opposite lane, and minimizing the damage to the occupants. According to its longitudinal position in the highway, it can be divided into roadbed guardrails and bridge guardrails; according to its horizontal position in the road, it can be divided into roadside guardrails and central divider guardrails; according to the degree of deformation after collision, can be divided It is a rigid guardrail, a semi-rigid guardrail and a flexible guardrail.

Central divider with guardrail: A guardrail placed in the central divider of the road to prevent uncontrolled vehicles from entering the opposite lane through the central divider and protecting the structure within the central divider.

Land objects refer to roads and various tangible objects on the ground around the road (such as road facilities, plants, buildings, etc.).

The point cloud data, the point cloud data used to represent the feature in the point cloud data.

Ground point cloud data, part of the cloud data used to represent the ground (such as the road surface, the surface connected to the road, the water surface).

The low point object point cloud data, the point cloud data point in the point cloud data for indicating that the value from the ground is greater than the first threshold and less than the second threshold; wherein the first threshold is less than the second threshold.

Other object point cloud data, the point cloud data used in the point cloud data to indicate that the value from the ground is greater than or equal to the second threshold.

3D curve fitting, using a continuous curve to approximate or compare the 3D points in the point cloud data, so that as many 3D points as possible fit the distribution of a continuous 3D curve, such as on the continuous 3D curve or from the 3D The curve is closer, and the three-dimensional curve is the result of three-dimensional curve fitting based on point cloud data. The implementation of the present invention is described in detail below with reference to the accompanying drawings.

An embodiment of the present invention provides a point cloud data processing method. As shown in FIG. 2, the method mainly includes:

Step 201: classify the collected point cloud data of each frame to obtain low ground point cloud data.

As an optional implementation manner, the step of obtaining low ground point cloud data includes:

According to the position information in the cloud data of each frame, the low point point cloud data is obtained.

As an optional implementation manner, before the collecting the collected point cloud data of each frame, the method further includes:

Obtaining a point cloud data file within a preset time period;

Each frame point cloud data is extracted from the point cloud data file based on the location information, the angle information, and the time information.

In practical applications, the vehicle environment is collected by setting a collection unit (such as a laser scanner, a three-dimensional camera), real-time positioning is performed by the positioning unit at each position in the traveling, and the environment is multi-angled by the collecting unit (for example) An acquisition of 0 to 360 full angles, an optional data structure for collecting point cloud data collected from any angle at each location is: geographic location, acquisition angle, three-dimensional point coordinates, and three-dimensional point attribute information.

Specifically, since the scanning device such as the laser scanner adopts a 360-degree rotational scanning mode, the point cloud obtained by rotating the laser scanner from 0 degrees to 360 degrees can be referred to as single-frame laser scanning line data. In the actual collection process, each scan line data is continuously stored to form a point cloud data file. Therefore, after obtaining the point cloud data, it is necessary to extract the point cloud data of each frame according to the angle information of each point (ie, the angle value is Point between 0-360 degrees). A typical single frame laser scan line data is shown in Figure 3.

As an optional implementation manner, the collected point cloud data of each frame is classified to obtain low ground point cloud data, including:

Determining, by the distance from each point in the cloud data of each frame point to the ground plane, a point whose distance value is greater than the first threshold and less than the second threshold as a low object point; wherein the first threshold is smaller than the second threshold .

Exemplarily, the single-frame point cloud data obtained in step 201 is extracted by a random sampling consensus algorithm (RANSAC, Random Sample Consensus) to obtain a ground plane equation a*x+b*y+c*z+d=0. , where c>0. Then, the point cloud coarse classification is performed according to the distance from each point in the single-frame point cloud data to the ground plane. The specific classification rules are as follows:

1) Calculate the distance from the point to the plane dist=|a*x+b*y+c*z+d|/sqrt(a*a+b*b+c*c);

2) When dist<= the first threshold dThred1, the point is judged as a ground point;

3) when dist> first threshold dThred1 and dist<second threshold dThred2, the point is judged as a low object point; wherein dThred1 is smaller than dThred2;

4) When dist ≥ the second threshold dThred2, the point is judged as another point.

Here, dThred1 can take a value of 0.3 meters and dThred2 takes a value of 1.5 meters. It should be noted that the dThred1 and dThred2 can be adaptively adjusted according to the extraction accuracy requirement.

Step 202: Extract candidate fence point cloud data from the low object point cloud data along a track direction.

As an optional implementation manner, the step of extracting candidate guardrail point cloud data from the low ground point cloud data includes:

The candidate guardrail point cloud data is extracted from the low ground point cloud data according to the distance between the low ground object and the specific side of the vehicle along the direction of the vehicle track.

As an optional implementation manner, the extracting, by the vehicle track, the candidate guardrail point cloud data from the low ground point cloud data, including:

The low point object cloud data in a certain range of the preset direction is extracted from the low point point cloud data along the vehicle track to obtain the candidate guard point cloud data.

Here, the preset direction may be the left side or the right side of the vehicle traveling direction.

In general, the predetermined direction is specifically the left side or the right side, depending on the position of the steering wheel of the vehicle in the vehicle. When the steering wheel is located on the left side of the vehicle, the preset direction is generally the left side; when the steering wheel is located on the right side of the vehicle, the preset direction is generally the right side.

Generally speaking, since the central separation zone of the road is generally between 0.5m and 1.5m from the ground, it falls right into the "low dwarf point" classification point; in addition, because China is driving right, the central separation zone of the road The guardrail is generally located to the left of the driving track. Therefore, the low datum points in the multi-frame point cloud data are accumulated to obtain a complete set of low-lying object point clouds; then, the vehicle trajectory is taken within a certain distance from the left side of the vehicle (vertical track trajectory) (for example) A point with a value of 15 meters) as a candidate guardian point cloud dataset.

Step 203: Perform spatial clustering on the candidate guardrail point cloud data to obtain a set of candidate guardrail point cloud.

As an implementation manner, the spatial clustering of the candidate guardrail point cloud data includes but is not limited to the following manners:

Space clustering of candidate guardrail point cloud data based on Euclidean distance clustering method;

Spatial clustering of candidate guardrail point cloud data based on graph theory-based clustering method;

Feature-based clustering method spatial clustering of candidate guardrail point cloud data.

As an optional implementation manner, the spatial clustering of the candidate guardrail point cloud data is performed to obtain a set of candidate guardrail point cloud clouds, including:

Extracting spatial features from candidate guard point cloud data of each frame;

By comparing the same spatial features of the candidate guardian point cloud data of each frame, clustering the candidate guardian point cloud data according to the same spatial feature of each frame candidate guardian point cloud data to form a plurality of candidate guardrails The point cloud collection, each candidate barrier point cloud collection includes a plurality of three-dimensional points and attribute information of the corresponding three-dimensional points.

Among them, the attribute information of the corresponding three-dimensional point, such as red, green and blue (RGB) color information, or reflection intensity information (Intensity).

Step 204: Identify each candidate guardrail point cloud set obtained by clustering, and obtain point cloud data of the road center divider guardrail.

As an implementation manner, the step of performing at least spatial feature recognition on each candidate barrier point cloud set obtained by clustering includes:

At least the spatial distribution feature and the linear feature recognition are performed on each candidate barrier point cloud set obtained by clustering.

As an optional implementation manner, the candidate guardrail point cloud set obtained by the cluster is identified, and the interference point cloud set is eliminated, and the point cloud data of the road center partition guardrail is obtained.

As an optional implementation manner, the candidate point cloud set obtained by clustering is identified, and the point cloud data of the road center partition guardrail is obtained, including:

Determining spatial distribution characteristics of each candidate point cloud set in three-dimensional space;

Determining linear features of each candidate point cloud set on a two-dimensional plane;

Determining spatial topological features between each candidate point cloud set on a two-dimensional plane after vertical projection and the ground;

The candidate guardrail point cloud sets are identified by combining the spatial distribution features in the three-dimensional space, the linear features on the two-dimensional plane, and the spatial topological features.

As an implementation manner, the spatial distribution feature combined in a three-dimensional space, the linear feature on the two-dimensional plane, and the spatial topological feature identify each candidate barrier point cloud set, including:

Extracting a first type of candidate point cloud set that satisfies a planar feature in a three-dimensional space and exhibits a continuous linear distribution feature on a two-dimensional plane after vertical projection;

Determining a spatial topological feature between the first type of candidate point cloud set on the two-dimensional plane after vertical projection and the ground;

A second type of candidate point cloud set that satisfies the spatial topological feature as an inclusion relationship is selected from the first type of candidate point cloud set, and the road center separation zone guard point cloud data is obtained based on the second type of candidate point cloud set.

Illustratively, the shape feature-based recognition method mainly considers that the road center divider guard rail presents a planar feature in three-dimensional space, as shown in FIG. 4; and on the two-dimensional plane after vertical projection, a continuous line is presented. Shape distribution characteristics, as shown in Figure 5. Therefore, this feature can be used to quickly remove the point cloud data (such as stationary vehicles on the road surface, anti-collision piers, and vegetation on both sides of the road).

The three-dimensional planar feature extraction may use a principal component analysis method (PCA) to calculate a spatial distribution feature of the candidate point cloud set;

Through the PCA algorithm, three large-to-small eigenvalues λ1, λ2, and λ3 of the point cloud distribution can be obtained, and the following spatial distribution characteristics are calculated:

P1=λ1/λ2;

P2=λ2/λ3;

In the above formula, P1 and P2 represent spatial distribution characteristics. When the point cloud is distributed in a body shape, λ1>λ2>λ3, the three eigenvalues are close to each other, and both P1 and P2 are small; when the point cloud is distributed in a plane, λ1> Λ2>>λ3, P1 is smaller, P2 is larger; when the point cloud is linearly distributed, λ1>>λ2>λ3, P1 is larger, and P2 is smaller.

It should be noted that ">>" is much greater than the meaning.

The calculation method of the linear features on the two-dimensional plane may adopt the Hough transform method, or first extract the two-dimensional image gradient, and then perform the line segment tracking method.

The identification method based on spatial topological features mainly considers the spatial topological relationship between the guardrail of the central divider of the road and the ground. Since most of the central dividers of the roads are in the form of fences, they can penetrate the point cloud. Therefore, in the two-dimensional plane after vertical projection, the spatial topology between them and the ground is inclusive, as shown in Figure 5. As shown in the figure; and the moving vehicle and the wall and other objects, the spatial topology between them and the ground is in a connected relationship.

In the actual extraction process, combined with the identification method of shape and spatial topological features, the extraction accuracy of the point cloud of the central divider of the road can be greatly improved, and the risk of false rejection can be reduced.

Step 205: Perform three-dimensional curve fitting on the point cloud data of the road center divider with the guardrail to obtain the road center divider guardrail data represented in the high-precision map.

Here, the amount of point cloud data of the central divider of the road extracted in step 204 is still large, and cannot be directly used for high-precision map representation. Therefore, the point cloud data extracted in step 204 is subjected to curve fitting by using a three-dimensional curve fitting method to obtain a final road guardrail three-dimensional curve data; wherein, a three-dimensional curve diagram of the road center divider with guardrail data is shown in FIG. 6.

As an embodiment, the three-dimensional curve fitting of the point cloud data of the central divider of the road is performed, including but not limited to the following manners:

A three-dimensional curve fitting method based on a polynomial equation for a three-dimensional curve fitting of a point cloud data of a road center divider with a guardrail;

The curve fitting method based on the random sampling consensus algorithm performs three-dimensional curve fitting on the point cloud data of the road center divider with guardrail.

As an optional implementation manner, the three-dimensional curve fitting of the point cloud data of the central divider of the road is performed, and the road central divider guardrail data represented in the high-precision map is obtained, including:

Performing a three-dimensional curve fitting on the point cloud data of the central divider of the road, and filtering out the point cloud data corresponding to the three-dimensional point that does not conform to the fitted three-dimensional curve;

The three-dimensional modeling is performed based on the point cloud data conforming to the features of the central divider of the road, forming a three-dimensional solid figure of the road center divider with a guardrail for rendering in a high-precision map.

Through the above scheme, the automatic extraction of the guard data of the road center separation zone based on the vehicle laser point cloud can be realized, and the high-precision road center separation zone guardrail data can be obtained.

At present, there is no distinction between the central divider of the road and the extraction method of the guardrails on both sides of the road. In the actual application of the driverless, the application requirements of the two types of guardrails are not the same and need to be distinguished. The application can quickly extract the road central divider guardrail data from the vehicle laser point cloud, and can provide basic data for high-end applications such as vehicle assisted positioning and driverless driving. The proposal of the present application can greatly improve the automatic extraction efficiency of the guardrail of the central divider of the road, reduce the workload of manual work, and reduce the production cost of the high-precision map.

The method for extracting the guardrail of the central divider of the road according to the embodiment of the present invention has the following advantages:

(1) By rough classification of single-frame point cloud data, and quickly locate the candidate point cloud data set of the guardrail, the amount of subsequent processing data can be greatly reduced, and the processing efficiency is improved;

(2) Using multi-frame laser point cloud fusion data to extract and identify guardrail data, which is more robust than single frame data;

(3) In the identification process, not only the shape characteristics of the guardrail of the central divider of the road are considered, but also the topological relationship between it and other features can be used to eliminate most of the disturbing features (such as vehicles and central flower beds). ), the recognition and extraction results are more robust.

The embodiment of the invention further provides a point cloud data processing method, comprising the steps of:

Obtaining low-land point cloud data based on at least position information in each frame point cloud data;

Extracting candidate guardrail point cloud data from the low ground point cloud data according to a distance between the low ground object and a specific side of the vehicle along a direction of the vehicle track;

Spatial clustering of candidate guardrail point cloud data according to preset conditions, to obtain a set of candidate guardrail point clouds;

At least spatial feature recognition is performed on each candidate barrier point cloud set obtained by clustering, and point cloud data of the road center partition guardrail is obtained;

Three-dimensional curve fitting is performed on the point cloud data of the central divider of the road, and the road center divider guardrail data represented in the high-precision map is obtained.

Based on the point cloud data processing method described above, FIG. 7 shows an optional logical function structure diagram of the point cloud data processing apparatus 10. The point cloud data processing apparatus 10 includes: a classification unit 21, a first extraction unit 22, and a clustering unit. 23. Identification unit 24 and fitting unit 25, each unit will be described below.

The classification unit 21 is configured to classify the collected point cloud data to obtain low point point cloud data;

The first extracting unit 22 is configured to extract candidate guardrail point cloud data from the low object point cloud data in a direction of the vehicle track;

The clustering unit 23 is configured to perform spatial clustering on the candidate guardrail point cloud data to obtain a set of candidate guardrail point cloud sets;

The identifying unit 24 is configured to identify each candidate guard point cloud set obtained by clustering, and obtain point cloud data of the road center partition guardrail;

The fitting unit 25 is configured to perform three-dimensional curve fitting on the road center separation barrier point cloud data, and obtain road central divider guardrail data represented in the high-precision map.

Further, the device further includes:

The second extraction unit 26 is configured to:

Obtaining a point cloud data file within a preset time period;

Each frame point cloud data is extracted from the point cloud data file based on the location information, the angle information, and the time information.

As an optional implementation manner, the classification unit 21 is specifically configured to:

Determining, by the distance from each point in the cloud data of each frame point to the ground plane, a point whose distance value is greater than the first threshold and less than the second threshold as a low object point; wherein the first threshold is smaller than the second threshold .

For example, the classification unit 21 integrates the received point cloud data into discrete point cloud data collected by the vehicle side at different locations and different acquisition angles, and integrates the received point cloud data into a form of “frames” for subsequent Processing, for example, for the received point cloud data, distinguishing the collection locations according to the labels of the geographic locations of the point cloud data, and for each collection location point cloud data, forming point cloud data of different acquisition angles of the corresponding locations into a corresponding position Frame point cloud data, each frame point cloud data includes coordinates and attribute information of a three-dimensional point obtained by collecting the road environment at different angles at corresponding positions.

As an optional implementation manner, the first extracting unit 22 is specifically configured to:

According to the direction of the vehicle track, the low point object cloud data in a certain range from the vehicle in the preset direction is extracted from the low point point cloud data, and the candidate fence point cloud data is obtained.

As an optional implementation manner, the clustering unit 23 is specifically configured to:

Extracting spatial features from candidate guard point cloud data of each frame;

By comparing the same spatial features of the candidate guardian point cloud data of each frame, clustering the candidate guardian point cloud data according to the same spatial feature of each frame candidate guardian point cloud data to form a plurality of candidate guardrails The point cloud collection, each candidate barrier point cloud collection includes a plurality of three-dimensional points and attribute information of the corresponding three-dimensional points.

As an optional implementation manner, the identifying unit 24 is specifically configured to:

Determining spatial distribution characteristics of each candidate point cloud set in three-dimensional space;

Determining linear features of each candidate point cloud set on a two-dimensional plane;

Determining spatial topological features between each candidate point cloud set on a two-dimensional plane after vertical projection and the ground;

The candidate guardrail point cloud sets are identified by combining the spatial distribution features in the three-dimensional space, the linear features on the two-dimensional plane, and the spatial topological features.

As an optional implementation, the identification unit 24 is further configured as follows:

Extracting a first type of candidate point cloud set that satisfies a planar feature in a three-dimensional space and exhibits a continuous linear distribution feature on a two-dimensional plane after vertical projection;

Determining a spatial topological feature between the first type of candidate point cloud set on the two-dimensional plane after vertical projection and the ground;

A second type of candidate point cloud set that satisfies the spatial topological feature as an inclusion relationship is selected from the first type of candidate point cloud set, and the road center separation zone guard point cloud data is obtained based on the second type of candidate point cloud set.

As an optional implementation, the fitting unit 25 is specifically configured to:

Performing a three-dimensional curve fitting on the point cloud data of the central divider of the road, and filtering out the point cloud data corresponding to the three-dimensional point that does not conform to the fitted three-dimensional curve;

The three-dimensional modeling is performed based on the point cloud data conforming to the features of the central divider of the road, forming a three-dimensional solid figure of the road center divider with a guardrail for rendering in a high-precision map.

Here, the purpose of screening the point cloud data corresponding to the three-dimensional point that does not conform to the fitted three-dimensional curve is to further reduce the noise in the extracted road facility point cloud data.

It should be understood by those skilled in the art that the functions of each unit in the point cloud data processing apparatus of this embodiment can be understood by referring to the related description of the foregoing point cloud data processing method.

In a practical application, the specific structures of the foregoing classification unit 21, the first extraction unit 22, the clustering unit 23, the identification unit 24, the fitting unit 25, and the first extraction unit 26 may all correspond to a processor. The specific structure of the processor may be a Central Processing Unit (CPU), a Micro Controller Unit (MCU), a Digital Signal Processing (DSP), or a programmable logic device (PLC). Programmable Logic Controller) A collection of electronic components or electronic components with processing functions. Wherein, the processor includes executable code, the executable code is stored in a storage medium, and the processor may be connected to the storage medium through a communication interface such as a bus, when performing a corresponding function of each unit Reading and running the executable code from the storage medium. The portion of the storage medium used to store the executable code is preferably a non-transitory storage medium.

The point cloud data processing device in this embodiment may be disposed on the vehicle side and the cloud server side.

The point cloud data processing apparatus provided by the embodiment of the present invention can be implemented in various manners, which will be exemplified below.

1) The point cloud data processing device is distributed on the cloud server side.

Referring to an optional hardware and software structure diagram of the point cloud data processing apparatus 10 shown in FIG. 8, the point cloud data processing apparatus 10 includes a component layer, an intermediate layer, an operating system layer, and a software layer. However, those skilled in the art should understand that the structure of the point cloud data processing apparatus 10 shown in FIG. 8 is merely an example and does not constitute a limitation on the structure of the point cloud data processing apparatus 10. For example, the point cloud data processing apparatus 10 sets more components than FIG. 8 according to implementation needs, or omits setting part components according to implementation needs.

The hardware layer of the point cloud data processing device 10 includes a processor 11, an input/output interface 13, a storage medium 14, a positioning module 12, a communication module 15, and an acquisition module 16; each component can communicate with the processor 11 via a system bus connection.

The processor 11 can be implemented by using a central processing unit (CPU), a microprocessor (MCU, Microcontroller Unit), an application specific integrated circuit (ASIC), or a Field-Programmable Gate Array (FPGA).

The input/output interface 13 can be implemented using input/output devices such as a display screen, a touch screen, and a speaker.

The storage medium 14 may be implemented by using a non-volatile storage medium such as a flash memory, a hard disk, or an optical disk, or may be implemented by using a volatile storage medium such as a double rate (DDR) double data rate cache, where the storage is useful to execute the point cloud data. An executable instruction that handles the method.

Illustratively, the storage medium 14 may be centrally located or distributed across different locations.

The communication module 15 provides the processor 11 with the access capability of the external data such as the storage medium 14 disposed off-site. For example, the communication module 15 can implement Near Field Communication (NFC) technology, Bluetooth technology, and purple. The short-range communication by the ZigBee technology can also implement communication systems such as Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), and its evolution system. Communication.

The collection module 16 is configured to perform multi-angle acquisition and output point cloud data, which may be implemented by a laser scanner or a three-dimensional camera. The point cloud data includes at least three-dimensional point coordinates, and the point cloud data according to the specific type of the collection module 16 further includes related The attribute information, such as the attribute information when the camera is a depth camera, is RGB information, and the attribute information is, for example, the reflection intensity information of the three-dimensional point (related to the gray scale) when the laser scanner is used.

The driver layer includes a middleware 17 for the operating system 18 to identify and communicate with the hardware layer components, such as a collection of drivers for the various components of the hardware layer.

The software layer includes providing a user with a high-precision map-based application such as a navigation application 19, and can also package various services based on high-precision maps into a callable application programming interface (API).

For example, when the communication module 15 establishes communication with the in-vehicle terminal in the vehicle, the software layer can provide a high-precision map-based service to the application in the in-vehicle terminal, including locating the current location of the vehicle, the navigation route query, and the like.

The point cloud data processing device is distributed on the cloud server side. A typical implementation scenario is shown in Figure 9-1. In the vehicle's driving environment, the point cloud data processing device sets the aforementioned acquisition module (such as a laser scanner) on the vehicle side. The point cloud data of different locations is collected at multiple angles (such as 0-360 degrees) along the road where the vehicle is traveling, and the label of the collection angle can be added for the collected point cloud data.

In addition, the point cloud data processing device may also be deployed with the foregoing positioning module in the vehicle side, and the positioning device is located in a real-time position of the vehicle based on a Global Positioning System (GPS), a Beidou satellite positioning navigation system, etc. Various forms of coordinate records), so that the collected point cloud data can be added to the collected geographical location label, and sent to the cloud server through the communication module deployed by the point cloud data processing device on the vehicle side, by the point cloud data processing device The processor of the server disposed in the cloud (by executing executable instructions in the storage medium) extracts point cloud data of the road facility from the point cloud data, and three-dimensionally models the road facility through the point cloud data of the road facility to form A three-dimensional solid figure of a road facility that can be used for rendering in a high-precision map.

2) The point cloud data processing device is distributed on the vehicle side.

An optional hardware and software structure diagram of the point cloud data processing device can still be seen in FIG. 8. The point cloud data processing device is distributed on the vehicle side. A typical implementation scenario is shown in FIG. 9-2, while the vehicle is running. The point cloud data processing device is provided with an acquisition module (such as a laser scanner) on the vehicle side to collect point cloud data at different positions (such as 0-360 degrees) to form different points of the point cloud data, which may be collected point clouds. The data adds a label for the acquisition angle.

In addition, the point cloud data processing device may also be provided with a positioning module in the vehicle side, and the positioning device locates the real-time position of the vehicle based on a global satellite positioning system (GPS), a Beidou satellite positioning navigation system, etc. (for example, using various forms of coordinate recording) Adding a label of the geographic location to the collected point cloud data and extracting the point cloud data of the road facility from the point cloud data by the controller provided by the point cloud data processing device on the vehicle side, the point cloud data pair of the road facility The road facilities are three-dimensionally modeled to form road facilities that can be used for rendering in high-precision maps. The point cloud data of the extracted road facilities can be sent to the cloud server, and the cloud server provides services based on the high-precision map of the road facilities. .

Hereinafter, the distribution of the point cloud data processing device on the vehicle side will be described as an example, and the description of the scene of the point cloud data processing shown in FIGS. 9-1 and 9-2 will be continued. Illustratively, FIG. 10 shows an alternative flow diagram of a point cloud data processing method. As shown in FIG. 10, the flow mainly includes:

In step 301, the road environment is collected when each vehicle travels along the road.

As mentioned above, the vehicle environment is collected by setting an acquisition module (such as a laser scanner, a three-dimensional camera), real-time positioning is performed by the positioning module at each position in the traveling, and the environment is multi-angled through the acquisition module ( For example, the acquisition of all angles from 0 to 360, an optional data structure for collecting point cloud data collected from any angle at each location is: geographic location, acquisition angle, three-dimensional point coordinates, and three-dimensional point attribute information.

Step 302: Each vehicle sends the point cloud data collected along the road to the cloud server side with the extraction function of the road center divider guardrail data.

In an embodiment, each vehicle can send the point cloud data collected by the collection module to the cloud server in real time through the set communication module, and the cloud server with high computing capability extracts the point cloud corresponding to the road facility from the point cloud data as soon as possible. Data for applications that require real-time updates to high-precision maps.

In another embodiment, each vehicle may send the point cloud data collected by the collection module to the cloud server when the predetermined transmission condition is reached, for the cloud server to extract the point cloud data corresponding to the road facility from the received point cloud data. Suitable for non-real-time application scenarios that require conditional updates to high-precision maps.

Illustratively, each vehicle may send point cloud data collected in a corresponding time period to the cloud server when the predetermined time (which may be periodic or non-periodic) arrives, for example, sending the collection every 5 minutes. Point cloud data.

Illustratively, each vehicle may transmit point cloud data collected at a corresponding mileage when the mileage traveled meets a predetermined mileage, for example, point cloud data to be collected within 1 km per 1 km travel is transmitted to the cloud server.

Step 303: The cloud server extracts each frame point cloud data according to the angle information of each point as needed.

The point cloud data received by the cloud server is discrete point cloud data collected at different locations and different collection angles. Here, the server integrates the received point cloud data into a "frame" format for subsequent processing, wherein Each acquisition location corresponds to at least one frame of point cloud data, and the number of frames of point cloud data formed for each location depends on the dwell time at that location and the speed at which the road environment is scanned during acquisition.

The cloud server receives point cloud data collected at various angles (0 to 360 degrees) at different locations, and for the received point cloud data, the collection location is distinguished according to the geographical location of the point cloud data, for each collection location point The cloud data forms point cloud data of different acquisition angles of corresponding positions to form a frame point cloud data of the corresponding position, and each frame point cloud data includes coordinates and attribute information of the three-dimensional points obtained by collecting the road environment at different angles at corresponding positions. .

Assuming that one frame of point cloud data of position 1 is formed, point cloud data having a tag of position 1 is first extracted from the received point cloud data, and point cloud data of a tag having position 1 is collected according to the point cloud data of each point cloud data. Arrange sequentially to form a corresponding one-frame point cloud data.

For example, an optional data structure of a frame of point cloud data at position 1 is (position 1, acquisition angle 0 - 3D point 1 coordinate - 3D point 1 attribute information; acquisition angle 1 - 3D point n coordinate - 3D point n attribute Information; ... acquisition angle 360 - 3D point 1 coordinates - 3D point 1 attribute information; acquisition angle 360 - 3D point n coordinates - 3D point n attribute information).

Step 304: The cloud server classifies the collected point cloud data of each frame to obtain low ground point cloud data.

In one embodiment, a plane equation corresponding to the ground plane is established according to the coordinates of the three-dimensional point of the point cloud data of each frame, and the height of each three-dimensional point in the frame point cloud data relative to the ground plane is obtained according to the plane equation, and the ground and the ground are The value range of the height corresponding to the object is divided into at least point cloud data (ground point cloud data) corresponding to the ground and point cloud data (ground point cloud data) corresponding to the ground object. Of course, it can also be divided into other types of point cloud data (referred to as other point cloud data) above the height of the feature.

For example, road facilities on the ground level differ in height from other plants, such as plants. The height of traffic lights is more than 1 meter, and the height of road barriers is generally between 0.3 and 1 meter. Plants near the road. Generally, flowers or other low-lying plants are generally below 0.3 meters.

In this way, by the height of the three-dimensional point of the cloud data of each frame relative to the ground plane, it can be preliminarily determined that the three-dimensional point is a corresponding ground plane, a corresponding feature or a corresponding higher object, and the three-dimensional point is divided into corresponding category point cloud data. Among them, depending on the road facilities, the point cloud data types that are highly adapted to the road facilities are also different, for example, the following may be included:

Case 1) For the road guardrail, since the height is in the range of the height of the feature, the feature point cloud data including the road guardrail is obtained by classification to the height of the road guardrail.

Case 2) For traffic lights, because the height is higher than the height range of the feature, the other point cloud data including the traffic lights and the height of the traffic lights are obtained by classification.

It can be seen that after classifying the point cloud data of each frame, the point cloud data categories that are highly adapted to the road facilities can be processed later, and the other types of point cloud data are filtered out (the subsequent processing is not necessary). This achieves the effect of preliminary screening of point cloud data including road facilities, reducing the amount of subsequent data processing.

Step 305: The cloud server extracts candidate guardrail point cloud data from the low object point cloud data along the vehicle track.

Generally speaking, the guardrail of the central divider of the road is generally 0.5m to 1.5m above the ground, which falls right into the "low-lying object point" classification point. For example, a point within a certain distance (for example, 15 meters) from the left side of the vehicle (vertical track trajectory) is taken along the vehicle trajectory as a candidate guard point cloud data set.

Exemplarily, for comprehensively determining whether the candidate point cloud data set belongs to the road center divider guard point cloud data based on the spatial distribution features of each segment candidate point cloud data set, the following manner may be adopted:

A method for judging a predetermined spatial distribution characteristic of a point cloud data of a guardrail based on a road center partition.

For example, dividing a certain candidate point cloud data set into a plurality of segments, and determining whether the feature value of the spatial distribution feature based on the plurality of segments conforms to a predetermined spatial distribution feature of the road center separation barrier point cloud data, such as whether or not The predetermined feature value of the central divider with guardrail point cloud data is consistent, or is in a predetermined range of values. If more than half of the segments meet the predetermined spatial distribution feature of the road center divider guardrail point cloud data, the candidate point cloud data set For the central separation of the road with guardrail point cloud data.

Step 306: The cloud server spatially clusters the candidate guardrail point cloud data to obtain a set of candidate guardrail point cloud.

Among them, the spatial clustering method includes, but is not limited to, a clustering method based on Euclidean distance, a clustering method based on graph theory, and a clustering method based on features.

Step 307: The cloud server identifies each candidate fence point cloud set obtained by the clustering, and removes the interference point cloud set to obtain point cloud data of the road center partition guardrail.

The cloud server combines the spatial distribution features in the three-dimensional space, the linear features on the two-dimensional plane, and the spatial topological features to identify the candidate guardrail point cloud sets. Combining the identification methods of shape and spatial topological features can greatly improve the extraction accuracy of the point cloud of the central divider of the road and reduce the risk of false rejection.

Step 308: The cloud server performs three-dimensional curve fitting on the road center separation barrier point cloud data, and obtains the road center separation zone guardrail data represented in the high-precision map.

In one embodiment, the cloud server performs a three-dimensional curve fitting on the road center separation barrier point cloud data, and filters the point cloud data corresponding to the three-dimensional point that does not conform to the fitted three-dimensional curve; The feature point cloud data is three-dimensionally modeled to form a three-dimensional solid figure of the road center divider with a guardrail for rendering in a high precision map.

The embodiment of the present invention further provides a computer storage medium, which may be a computer readable storage medium included in the memory in the above embodiment, or may be separately readable by a computer that is not assembled into the terminal. Storage medium. The computer readable storage medium stores one or more computer executable instructions that are used by one or more processors to perform the community discovery method of embodiments of the present invention. Specifically, the computer executable instructions are configured to perform: classifying the collected point cloud data to obtain low ground point cloud data; and in the direction of the vehicle track, from the low ground point cloud The candidate guardrail point cloud data is extracted from the data; the candidate guardrail point cloud data is spatially clustered to obtain the candidate guardrail point cloud set; the candidate guardrail point cloud set obtained by the cluster is identified, and the point of the road center divider guardrail is obtained. Cloud data; three-dimensional curve fitting is performed on the point cloud data of the central divider of the road, and the road center divider guardrail data represented in the high-precision map is obtained.

As an implementation manner, the computer executable instructions are configured to: before collecting the collected point cloud data of each frame, acquiring a point cloud data file in a preset time period; according to the location information, the angle information, and The time information extracts each frame point cloud data from the point cloud data file.

In one embodiment, the computer executable instructions are configured to: determine, according to a distance from each point in the cloud data of each frame point to a ground plane, a point whose distance value is greater than a first threshold and less than a second threshold as a low ground An object point; wherein the first threshold is less than the second threshold.

In one embodiment, the computer executable instructions are configured to: extract, in the direction of the track of the vehicle, from the low point object cloud data, a low ground object in a certain range from a side of the preset direction Point cloud data to get candidate fence point cloud data.

As an embodiment, the computer executable instructions are configured to: extract spatial features from candidate guard point cloud data of each frame; and compare the same spatial features of each frame candidate point cloud data by using Each frame candidate cluster point cloud data has the same spatial feature, and each frame candidate guard point cloud data is clustered to form a plurality of candidate guardrail point cloud sets, each candidate guardrail point cloud set includes a plurality of three-dimensional points and corresponding The attribute information of the 3D point.

In one embodiment, the computer executable instructions are configured to: determine spatial distribution features of each candidate point cloud set in three-dimensional space; determine linear features of each candidate point cloud set on a two-dimensional plane; determine each candidate The spatial topological features of the point cloud set on the two-dimensional plane after vertical projection and the ground; the spatial distribution features in the three-dimensional space, the linear features on the two-dimensional plane, and the spatial topological features of the candidate guardrail point clouds The collection is identified.

In one embodiment, the computer executable instructions are configured to: extract a first type of candidate point that satisfies a planar feature in a three-dimensional space and a continuous linear distribution feature on a two-dimensional plane after vertical projection a cloud set; determining a spatial topological feature between the first type of candidate point cloud set on a two-dimensional plane after vertical projection and the ground; and selecting, from the first set of candidate point cloud sets, a second satisfying spatial topological feature as an inclusion relationship A class candidate point cloud set, based on the second type of candidate point cloud sets, obtains a road center divider with guardrail point cloud data.

In one embodiment, the computer executable instructions are configured to: perform a three-dimensional curve fitting on the road center divider fence point cloud data, and filter out point cloud data corresponding to the three-dimensional point that does not conform to the fitted three-dimensional curve. 3D modeling based on point cloud data conforming to features of the central divider of the road, forming a three-dimensional solid figure of the road center divider with a guardrail for rendering in a high precision map.

In the several embodiments provided by the present invention, it should be understood that the disclosed method and apparatus may be implemented in other manners. The device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be another division manner, for example, multiple modules or components may be combined, or Can be integrated into another system, or some features can be ignored or not executed. In addition, the communication connections between the various components shown or discussed may be indirect coupling or communication connections through some interfaces, devices or modules, and may be electrical, mechanical or otherwise.

The modules described above as separate components may or may not be physically separated. The components displayed as modules may or may not be physical modules, that is, may be located in one place or distributed to multiple network modules; Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may be separately used as one module, or two or more modules may be integrated into one module; The module can be implemented in the form of hardware or in the form of hardware plus software function modules.

It can be understood by those skilled in the art that all or part of the steps of implementing the above method embodiments may be completed by hardware related to program instructions, and the foregoing program may be stored in a computer readable storage medium, and the program is executed when executed. The foregoing storage medium includes: a mobile storage device, a random access memory (RAM), a read-only memory (ROM), a magnetic disk, or an optical disk. A medium that can store program code.

Alternatively, the above-described integrated unit of the present invention may be stored in a computer readable storage medium if it is implemented in the form of a software function module and sold or used as a standalone product. Based on such understanding, the technical solution of the embodiments of the present invention may be embodied in the form of a software product in essence or in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for making A computer device (which may be a personal computer, server, or network device, etc.) performs all or part of the methods described in various embodiments of the present invention.

The point cloud data processing method and apparatus described in the embodiments of the present invention are only exemplified by the above embodiments, but are not limited thereto, and those skilled in the art should understand that the technical solutions described in the foregoing embodiments can still be performed. Modifications, or equivalents to some or all of the technical features, and the modifications or substitutions do not depart from the scope of the technical solutions of the embodiments of the present invention.

The above is only a specific embodiment of the present invention and is not intended to limit the scope of the present invention.

Industrial applicability

The technical solution of the embodiment of the present invention obtains the low point object point cloud data first; extracts the candidate guardrail point cloud data from the low ground object point cloud data along the vehicle track; and performs spatial clustering on the candidate guardrail point cloud data. Obtaining a set of candidate guardrail point clouds; identifying each candidate guardrail point cloud set obtained by clustering, obtaining point cloud data of the road center divider guardrail; and performing three-dimensional curve fitting on the road center divider guardrail point cloud data Obtaining the data of the central divider of the road in the high-precision map; thus, it is possible to quickly extract the data of the central divider of the road from the on-board laser point cloud, thereby providing high-end applications such as vehicle-assisted positioning and driverless driving. The basic data can greatly improve the automatic extraction efficiency of the guardrail on the central divider of the road, reduce the workload of manual work, and reduce the production cost of high-precision maps.

Claims (21)

A point cloud data processing method, comprising: Sorting the collected point cloud data to obtain low-point point cloud data; Extracting candidate guardrail point cloud data from the low ground point cloud data in a direction of the vehicle track; Spatial clustering of candidate guardrail point cloud data to obtain a set of candidate guardrail point cloud sets; Identifying each candidate fence point cloud set obtained by clustering, and obtaining point cloud data of the road center divider with guardrail; Three-dimensional curve fitting is performed on the point cloud data of the central divider of the road, and the road center divider guardrail data represented in the high-precision map is obtained. The method of claim 1 wherein said step of obtaining low ground point cloud data comprises: According to the position information in the cloud data of each frame, the low point point cloud data is obtained. The method of claim 1 wherein said step of extracting candidate barrier point cloud data from said low landmark point cloud data comprises: The candidate guardrail point cloud data is extracted from the low ground point cloud data according to the distance between the low ground object and the specific side of the vehicle along the direction of the vehicle track. The method according to claim 1, wherein the step of performing at least spatial feature recognition on each candidate barrier point cloud set obtained by clustering comprises: At least the spatial distribution feature and the linear feature recognition are performed on each candidate barrier point cloud set obtained by clustering. The method according to claim 1, wherein before the classifying the collected point cloud data, the method further comprises: Obtaining a point cloud data file within a preset time period; Each frame point cloud data is extracted from the point cloud data file based on the location information, the angle information, and the time information. The method according to claim 1, wherein the classifying the collected point cloud data of each frame to obtain low ground point cloud data comprises: Determining, by the distance from each point in the cloud data of each frame point to the ground plane, a point whose distance value is greater than the first threshold and less than the second threshold as a low object point; wherein the first threshold is smaller than the second threshold . The method of claim 1 wherein said extracting candidate barrier point cloud data from said low object point cloud data along a vehicle trajectory comprises: According to the direction of the vehicle track, the low point object cloud data in a certain range from the vehicle in the preset direction is extracted from the low point point cloud data, and the candidate fence point cloud data is obtained. The method according to claim 1, wherein the spatial clustering of the candidate guardrail point cloud data is performed to obtain a set of candidate guardrail point cloud clouds, including: Extracting spatial features from candidate guard point cloud data of each frame; By comparing the same spatial features of the candidate guardian point cloud data of each frame, clustering the candidate guardian point cloud data according to the same spatial feature of each frame candidate guardian point cloud data to form a plurality of candidate guardrails The point cloud collection, each candidate barrier point cloud collection includes a plurality of three-dimensional points and attribute information of the corresponding three-dimensional points. The method according to claim 1, wherein the identifying the set of candidate guardrail point cloud obtained by the clustering, and obtaining the road central separation barrier point cloud data, comprising: Determining spatial distribution characteristics of each candidate point cloud set in three-dimensional space; Determining linear features of each candidate point cloud set on a two-dimensional plane; Determining spatial topological features between each candidate point cloud set on a two-dimensional plane after vertical projection and the ground; Combining the spatial distribution features in the three-dimensional space, the linear features on the two-dimensional plane, and the spatial topological features, the candidate guardrail point cloud sets are identified. The method according to claim 9, wherein the spatial distribution feature combined in three-dimensional space, the linear feature on the two-dimensional plane, and the spatial topological feature identify each candidate barrier point cloud set, including: Extracting a first type of candidate point cloud set that satisfies a planar feature in a three-dimensional space and exhibits a continuous linear distribution feature on a two-dimensional plane after vertical projection; Determining a spatial topological feature between the first type of candidate point cloud set on the two-dimensional plane after vertical projection and the ground; A second type of candidate point cloud set that satisfies the spatial topological feature as an inclusion relationship is selected from the first type of candidate point cloud set, and the road center separation zone guard point cloud data is obtained based on the second type of candidate point cloud set. The method according to claim 1, wherein said three-dimensional curve fitting is performed on said road center divider guardrail point cloud data, and obtaining road central divider guardrail data represented in a high-precision map, comprising: Performing a three-dimensional curve fitting on the point cloud data of the central divider of the road, and filtering out the point cloud data corresponding to the three-dimensional point that does not conform to the fitted three-dimensional curve; The three-dimensional modeling is performed based on the point cloud data conforming to the features of the central divider of the road, forming a three-dimensional solid figure of the road center divider with a guardrail for rendering in a high-precision map. A point cloud data processing method, comprising the steps of: Obtaining low-land point cloud data based on at least position information in each frame point cloud data; Extracting candidate guardrail point cloud data from the low ground point cloud data according to a distance between the low ground object and a specific side of the vehicle along a direction of the vehicle track; Spatial clustering of candidate guardrail point cloud data according to preset conditions, to obtain a set of candidate guardrail point cloud clouds; At least spatial feature recognition is performed on each candidate barrier point cloud set obtained by clustering, and point cloud data of the road center partition guardrail is obtained; Three-dimensional curve fitting is performed on the point cloud data of the central divider of the road, and the road center divider guardrail data represented in the high-precision map is obtained. A point cloud data processing device includes: The classification unit is configured to classify the collected point cloud data to obtain low point point cloud data; a first extracting unit configured to extract candidate guardrail point cloud data from the low object point cloud data along a direction of the vehicle track; a clustering unit configured to spatially cluster candidate point cloud data to obtain a set of candidate guardrail point clouds; The identification unit is configured to identify each candidate fence point cloud set obtained by the cluster, and obtain point cloud data of the road center divider guardrail; The fitting unit is configured to perform three-dimensional curve fitting on the road center divider fence point cloud data, and obtain the road center divider fence data represented in the high-precision map. The apparatus of claim 13, wherein the apparatus further comprises a second extraction unit configured to: Obtaining a point cloud data file within a preset time period; Each frame point cloud data is extracted from the point cloud data file based on the location information, the angle information, and the time information. The apparatus of claim 13 wherein said classification unit is configured to: Determining, by the distance from each point in the cloud data of each frame point to the ground plane, a point whose distance value is greater than the first threshold and less than the second threshold as a low object point; wherein the first threshold is smaller than the second threshold . The apparatus of claim 13, wherein the first extraction unit is configured to: According to the direction of the vehicle track, the low point object cloud data in a certain range from the vehicle in the preset direction is extracted from the low point point cloud data, and the candidate fence point cloud data is obtained. The apparatus of claim 13 wherein said clustering unit is configured to: Extracting spatial features from candidate guard point cloud data of each frame; By comparing the same spatial features of the candidate guardian point cloud data of each frame, clustering the candidate guardian point cloud data according to the same spatial feature of each frame candidate guardian point cloud data to form a plurality of candidate guardrails The point cloud collection, each candidate barrier point cloud collection includes a plurality of three-dimensional points and attribute information of the corresponding three-dimensional points. The apparatus of claim 13 wherein said identification unit is configured to: Determining spatial distribution characteristics of each candidate point cloud set in three-dimensional space; Determining linear features of each candidate point cloud set on a two-dimensional plane; Determining spatial topological features between each candidate point cloud set on a two-dimensional plane after vertical projection and the ground; Combining the spatial distribution features in the three-dimensional space, the linear features on the two-dimensional plane, and the spatial topological features, the candidate guardrail point cloud sets are identified. The apparatus of claim 18, wherein the identification unit is configured to: Extracting a first type of candidate point cloud set that satisfies a planar feature in a three-dimensional space and exhibits a continuous linear distribution feature on a two-dimensional plane after vertical projection; Determining a spatial topological feature between the first type of candidate point cloud set on the two-dimensional plane after vertical projection and the ground; A second type of candidate point cloud set that satisfies the spatial topological feature as an inclusion relationship is selected from the first type of candidate point cloud set, and the road center separation zone guard point cloud data is obtained based on the second type of candidate point cloud set. The apparatus of claim 13 wherein said fitting unit is configured to: Performing a three-dimensional curve fitting on the point cloud data of the central divider of the road, and filtering out the point cloud data corresponding to the three-dimensional point that does not conform to the fitted three-dimensional curve; The three-dimensional modeling is performed based on the point cloud data conforming to the features of the central divider of the road, forming a three-dimensional solid figure of the road center divider with a guardrail for rendering in a high-precision map. A computer storage medium having stored therein computer executable instructions for performing the point cloud data processing method of any one of claims 1 to 11, and/or claims The point cloud data processing method described in 12.

Images