CN109117847A - Component dividing method and device for vehicle damage identification - Google Patents
Component dividing method and device for vehicle damage identification Download PDFInfo
- Publication number
- CN109117847A CN109117847A CN201811012740.1A CN201811012740A CN109117847A CN 109117847 A CN109117847 A CN 109117847A CN 201811012740 A CN201811012740 A CN 201811012740A CN 109117847 A CN109117847 A CN 109117847A
- Authority
- CN
- China
- Prior art keywords
- component
- visual signature
- feature
- result
- damaged vehicle
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
This specification embodiment provides a kind of component dividing method and device for vehicle damage identification, according to this method embodiment, at least one photo site of damaged vehicle is obtained first, and then threedimensional model extracts the geometry feature of damaged vehicle from threedimensional model, and the visual signature of damaged vehicle is extracted from least one photo site, then, it is at least based on extracted visual signature and geometry feature, component segmentation is carried out to the damaged vehicle.The embodiment can make component segmentation result more acurrate, to improve the validity of vehicle damage identification.
Description
Technical field
This specification one or more embodiment is related to field of computer technology, more particularly to is damaged by computer in vehicle
Hurt the method and apparatus that component segmentation is carried out in identification process.
Background technique
Vehicle is the general name of various vehicles, such as may include bicycle, car, truck, lorry, train etc..Vehicle
In injury event, often because of damaged vehicle caused by unexpected or artificial mistake, such as vehicle is scraped, collision etc..Damaged vehicle
Subsequent processing often relates to the identification to vehicle damage, to provide foundation to vehicle maintenance, settlement of insurance claim etc..
In routine techniques, photo, the video etc. at scene are often utilized to the qualification process of vehicle damage, is identified from image
Damage.Such damnification recognition method often merely relates to the visual signature of damaged vehicle.However, for some more complex vehicles
Damage, the visual signature for only relying on image is less susceptible to accurately identify.Accordingly, it would be desirable to there is improved plan, using more
Vehicle data, more accurately carry out component segmentation, thus improve vehicle damage identification validity.
Summary of the invention
This specification one or more embodiment describes a kind of component segmentation side in vehicle damage identification process
Method and device can utilize the Image Visual Feature and three-dimensional structural feature of damaged vehicle, simultaneously to determine the portion of damaged vehicle
Part segmentation result keeps segmentation result more acurrate, to improve the validity of vehicle damage identification.
According in a first aspect, providing a kind of component dividing method for vehicle damage identification, comprising: obtain impaired vehicle
At least one photo site and threedimensional model；The geometry for extracting the damaged vehicle from the threedimensional model is special
Sign；The visual signature of the damaged vehicle is extracted from least one described photo site；At least it is based on the visual signature, institute
It states geometry feature and component segmentation is carried out to the damaged vehicle.
According to a kind of embodiment, described at least based on the visual signature, the geometry feature to described impaired
It includes: the disaggregated model that geometry feature input is trained in advance that vehicle, which carries out component segmentation, passes through the classification mould
Type exports the recognition result to component, to obtain primary segmentation result；Based on the visual signature to the primary segmentation knot
Fruit is modified, with generating unit segmentation result.
In one embodiment, described that the primary segmentation result is modified based on the visual signature, to generate
Component segmentation result includes: by the visual signature and the primary segmentation result Introduced Malaria model, according to the amendment mould
The output result of type determines the component segmentation result.
In another embodiment, described that the primary segmentation result is modified based on the visual signature, with life
It include: that the visual signature is mapped to the threedimensional model at component segmentation result；According to preset modification rule, in conjunction with reflecting
It penetrates result and adjusts the primary segmentation as a result, to obtain the component segmentation result, wherein the modification rule is for describing portion
The corresponding relationship of part partition error item and modification strategy.
According to another embodiment, it is described at least based on the visual signature, the geometry feature to it is described by
It includes: the parted pattern that the visual signature, geometry feature input is trained in advance that damage vehicle, which carries out component segmentation,
The component segmentation result is determined according to the output result of the parted pattern.
In one embodiment, the geometry feature includes at least one of the following: shape feature, volume characteristic, top
Point coordinate, curved surface normal angle, curved surface smoothness, unblind are apart from described function.
In one embodiment, the visual signature includes at least one of the following: that region links up feature, contour feature, face
Color consistency feature.
According to second aspect, a kind of component segmenting device for vehicle damage identification is provided, comprising: acquiring unit is matched
It is set at least one photo site and threedimensional model for obtaining damaged vehicle；First extraction unit is configured to from the three-dimensional
The geometry feature of damaged vehicle described in model extraction；Second extraction unit is configured to from least one described photo site
The middle visual signature for extracting the damaged vehicle；Cutting unit is configured at least based on the visual signature, the geometry
Feature carries out component segmentation to the damaged vehicle.
According to the third aspect, a kind of computer readable storage medium is provided, computer program is stored thereon with, when described
When computer program executes in a computer, enable computer execute first aspect method.
According to fourth aspect, a kind of calculating equipment, including memory and processor are provided, which is characterized in that described to deposit
It is stored with executable code in reservoir, when the processor executes the executable code, the method for realizing first aspect.
The component dividing method and device for vehicle damage identification provided by this specification embodiment, obtains first
Then at least one photo site and threedimensional model of damaged vehicle extract the geometry of damaged vehicle from threedimensional model
Feature, and it is special to be at least then based on extracted vision for the visual signature of extraction damaged vehicle from least one photo site
It seeks peace geometry feature, component segmentation is carried out to the damaged vehicle.In this way, the two-dimensional of damaged vehicle can be made full use of
Image Visual Feature and three-dimensional structural feature keep component segmentation result more acurrate, to improve the validity of vehicle damage identification.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, required use in being described below to embodiment
Attached drawing be briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for this
For the those of ordinary skill of field, without creative efforts, it can also be obtained according to these attached drawings others
Attached drawing.
Fig. 1 shows the implement scene schematic diagram of one embodiment of this specification disclosure；
Fig. 2 shows the component dividing method flow charts for vehicle damage identification according to one embodiment；
Fig. 3 shows the specific example of a three-dimensional vehicle model；
Fig. 4 shows the specific example that visual signature is mapped to threedimensional model；
Fig. 5 shows the schematic block diagram of the component segmenting device for vehicle damage identification according to one embodiment.
Specific embodiment
With reference to the accompanying drawing, the scheme provided this specification is described.
Fig. 1 is the implement scene schematic diagram of one embodiment that this specification discloses.In the implement scene, for impaired
Vehicle, computing platform can carry out component segmentation to it, to determine damaged parts, so that subsequent progress setting loss (determines damage
As a result, such as " bumper scraping "), determine the processes such as maintenance program, settlement of insurance claim.Wherein, computing platform can be with one
Determine electronic equipment of computing capability, such as desktop computer, laptop etc..The processes such as subsequent setting loss can be by other
Equipment is completed, and can also be completed by above-mentioned computing platform, this specification embodiment is not construed as limiting this.
Specifically, above-mentioned computing platform can obtain at least one photo site of damaged vehicle first, and three-dimensional
Model.Then, computing platform can extract the geometry feature of damaged vehicle from threedimensional model, at the same time it can also to
The visual signature of damaged vehicle is extracted in a few photo site.Wherein: geometry feature for example can be component shape spy
Sign, piece volumes, component vertex relative coordinate, curved surface normal angle, curved surface smoothness, unblind are apart from described function
The feature of (truncated signed distance function, TSDF) etc.；Visual signature for example can be area
Domain links up the feature of feature, component outline feature, colour consistency feature or the like.
At least based on above-mentioned visual signature, geometry feature, computing platform can carry out component point to damaged vehicle
It cuts.In embodiment on the one hand, computing platform can use parted pattern trained in advance, and the vision of damaged vehicle is special
Sign, geometry feature input the parted pattern, and the component point of damaged vehicle can be determined according to the output result of parted pattern
Cut result.In embodiment on the other hand, computing platform can also be using disaggregated model trained in advance, by damaged vehicle
Geometry feature inputs the disaggregated model, and the primary segmentation result of component is determined by the output result of disaggregated model.Then,
Visual signature is recycled to be modified the primary segmentation result, to obtain the final component segmentation result of damaged vehicle.
In this way, the threedimensional model of damaged vehicle can be made full use of, therefrom to extract the geometry feature of all parts,
In combination with the Image Visual Feature in photo site, the mesh that the more acurrate component to damaged vehicle is split can achieve
, to improve the validity of vehicle damage identification.The specific implementation procedure of above-mentioned scene is described below.
Fig. 2 shows the component dividing method flow charts for vehicle damage identification according to one embodiment.This method
Executing subject can be it is any there is calculating, the system of processing capacity, unit, platform or server, such as shown in Fig. 1
Computing platform etc..More specifically, such as can be the setting loss server for providing support for car damage identification service.
If Fig. 2 shows, method includes the following steps: step 21, obtains at least one photo site of damaged vehicle, and
Threedimensional model；Step 22, the geometry feature of damaged vehicle is extracted from above-mentioned threedimensional model；Step 23, existing from least one
The visual signature of damaged vehicle is extracted in field picture；Step 24, at least based on above-mentioned visual signature, geometry feature to impaired
Vehicle carries out component segmentation.
Firstly, obtaining at least one photo site and threedimensional model of damaged vehicle in step 21.It is appreciated that pair
For damaged vehicle, its photo site, such as smart phone, camera etc. can be obtained in several ways.Photo site
There can be one or more, according to actual needs or scene can obtain situation and determine.
Wherein, the threedimensional model of damaged vehicle can be acquired by field device, can also be based on the information of collection in worksite
It determines, this specification embodiment is not construed as limiting this.
In embodiment on the one hand, field device may include the equipment of laser, radar camera etc.By this kind of
Equipment can acquire point cloud chart of damaged vehicle, etc..Wherein, point cloud chart by by field device (measuring instrument) obtain by
The point data collection on damage vehicle appearance surface is combined into.The threedimensional model of damaged vehicle can be rebuild by point cloud chart, herein no longer
It repeats.
In embodiment on the other hand, the photo, video, depth map of damaged vehicle can be acquired by field device
Deng.By these data, the threedimensional model of damaged vehicle can be rebuild, can also determine the vehicle classification of damaged vehicle, with from
The corresponding threedimensional model of corresponding vehicle classification is searched in model data library.According to an optional implementation, from vehicle number
In the case where searching threedimensional model in library, it is also based on the information such as photo site and changes to the threedimensional model found
In generation, updates, to obtain the threedimensional model of more accurate damaged vehicle.
Then, by step 22, the geometry feature of damaged vehicle is extracted from above-mentioned threedimensional model.It is appreciated that by
The threedimensional model of vehicle is damaged before no progress component segmentation, often may include the feature of some overall profiles, such as shape
Feature in terms of shape, volume, curved surface, these features can be referred to as geometry feature.
In one embodiment, the geometry feature of damaged vehicle may include shape feature.Here shape feature
It can be Global shape feature, be also possible to local shape characteristics.As shown in figure 3, being the concrete example of a three-dimensional vehicle model
Son.Shape feature may include the profile of entire vehicle, be also possible to local configuration, such as four circles of vehicle bottom etc..
It is readily appreciated that, the shape of various pieces and component have much relations in threedimensional model, for example, four of vehicle bottom the reason is that wheel
A possibility that it is larger.Therefore, shape feature can be used for the identification and segmentation of component.
In another embodiment, the geometry feature of damaged vehicle can also include volume characteristic.Volume characteristic is
For indicating the feature of damaged vehicle size, size.Here, volume characteristic can be absolute volume feature, be also possible to opposite
Volume characteristic.For example, volume characteristic can be according to practical ruler in the case where the threedimensional model of damaged vehicle includes actual size
It is very little to calculate, maximum length, maximum height, maximum width can also be taken respectively as length, volume is expressed as " long
The form of × wide × height ", can also be everywhere highly distribution to indicate, etc..Relative volume can according to partial volume with it is whole
The ratio of a threedimensional model volume determines.Such as in Fig. 3, the volume of four circular portions of vehicle bottom and entire threedimensional model
The ratio of volume, it will be understood that relative volume does not need actual size, it is only necessary to relative size, therefore be easier to determine.When three
When the volume characteristic of some part of dimension module determines, the foundation of component segmentation can be used as.
In another embodiment, the geometry feature of damaged vehicle can also include apex coordinate.Here, vertex can
To be the spatial point for being constituted each known coordinate of threedimensional model.Apex coordinate can be each vertex relative to threedimensional model institute
In the coordinate of space coordinates origin.According to each apex coordinate, shape and position in conjunction with all parts can be by three-dimensional moulds
Type is divided into different components.
In another embodiment, the geometry feature of damaged vehicle can also include curved surface normal angle.It is easy reason
Solution, curved surface normal direction can be direction of the normal of each plane of each curved surface refinement under space coordinates, each for describing
The directional information of the plane of a refinement.Correspondingly, curved surface normal angle can be normal and the threedimensional model place of each plane
Angle formed by the reference axis of space coordinates.As it can be seen that the chamfered shape that curved surface normal angle can be used as identification component is related
Feature.
In another embodiment, the geometry feature of damaged vehicle can also include curved surface smoothness.In general, bent
Face smoothness can the indexs such as the continuous, continual curvature of, tangent line continuous by position measure.Position is continuous, i.e. two, section
The endpoint of line is overlapped；Tangent line is continuous, i.e. point of intersection tangent line is consistent, shows the smooth no dog-ear of plane；Continual curvature shows two songs
Line is tangent, their tangential radii is the same, the smooth no dog-ear of curved surface.As it can be seen that by curved surface smooth degree can be used as with
The relevant feature of the chamfered shape of identification component.
In another embodiment, the geometry feature of damaged vehicle can also include unblind distance description letter
Number.Unblind is apart from the function that described function is for describing truncation distance.According to camera coordinates system, camera photocentre is original
Point, taking camera optical axis is Z axis, and pixel coordinate axis u and camera coordinates axis X are in the same direction.Coordinate system meets right-handed system rule.For one
Depth image, each three-dimensional point of the back project into three-dimensional space correspond to each of former depth image pixel
Point regards the cloud as a three-dimension curved surface, is denoted as F (x, y, z)=0.Using the optical axis of camera as Z axis, for a three-dimensional point
(x0, y0, z0), the three-dimensional point to above-mentioned three-dimension curved surface apart from described function d=F (x0, y0, z0).When | d | when ﹥ t, d=0, t
Distance is exactly truncated.In each section of unblind apart from the threedimensional model that described function is used to describe to be mapped by depth map
Each point specific location.Therefore, it can be used as the foundation of component identification segmentation.
In more embodiments, the geometry feature of damaged vehicle can also include that other are relevant to component identification
Feature, this is no longer going to repeat them.
By the step, can be extracted from the threedimensional model of damaged vehicle, it is three-dimensional, three-dimensional to be identified about component
Feature, utilize the information of richer damaged vehicle.
In some possible embodiments, different part each in the threedimensional model of damaged vehicle can also be mentioned respectively
Take geometry feature.In addition, their spatial relation feature can also be extracted for various pieces.For example, two not
With relative positional relationship (such as relative coordinate), the neighbouring relations of each section at the center of part.The geometry of each section
Spatial relation feature between feature and each section can also be used as a part of the integrated structure feature of damaged vehicle.Such as
This, can use richer three-dimensional information and carry out component segmentation.
In addition, extracting the visual signature of damaged vehicle from least one photo site in step 23.It is appreciated that view
Feel that feature is often that vision is appreciable, visible feature.Such as the feature in embodiment color, boundary, gap etc..
In one embodiment, the visual signature of damaged vehicle may include the connectivity of region feature.Connected region
(Connected Component) generally refer in image with same pixel value and position it is adjacent foreground pixel point composition
Image-region.The connectivity of region feature can be indicate region whether be connected region, connected region type (simply connected region or
Duplicate connected domain), the feature on connected region boundary etc..It is appreciated that the corresponding part of connected region may correspond to the same portion
Part, other regions in a duplicate connected domain likely correspond to another component other than the corresponding component of the duplicate connected domain,
As other connected regions in duplicate connected domain car door correspond to door handle component, etc..
In another embodiment, the visual signature of damaged vehicle can also include contour feature.Profile is that composition is any
The boundary of one shape or trim line, can often define object range.Profile can be formed by gap, edge line etc..For example,
Gap between two car doors can separate two car doors.Contour feature can be by multiple on gap, edge line etc.
The curve that point indicates, such as { (x1, y1, z1), (x2, y2, z2)……}。
In having one embodiment, the visual signature of damaged vehicle can also include colour consistency feature.It is appreciated that
It is appreciated that the color of the same component is often with uniformity.Colour consistency feature can pass through rgb value, gray value etc.
It indicates.
In other embodiments, the visual signature of damaged vehicle can also include that other visions are appreciable, surface
Feature, this is no longer going to repeat them.
Worth explanation is that step 22 and step 23 are two mutually indepedent steps, and from different objects, (one is respectively
Threedimensional model.One is two-dimension picture) different features is extracted, therefore, the two steps can execute parallel, can also exchange
Sequence executes, and this specification embodiment is not construed as limiting this.
Step 24, component segmentation is at least carried out to damaged vehicle based on above-mentioned visual signature, geometry feature.It can manage
Solution, the geometry feature extracted by step 22 are to identify relevant feature with component, and the vision extracted by step 23 is special
Sign can be used for the more Accurate Segmentation of component, therefore, be based on above-mentioned visual signature and geometry feature, can be to impaired vehicle
Carry out component segmentation.
Component segmentation result can be described by photo site, such as be described by rectangle frame or edge line different
Component can also be described by threedimensional model, such as be partitioned into all parts in the three-dimensional model with three-dimensional frame or edge line,
It can also be described by the file of text or extended formatting, such as front left wheel { (x1, y1, z1), (x2, y2, z2) ..., etc.
Deng.The embodiment of this specification is not construed as limiting this.
According to a possible design, the disaggregated model that can first train the input of above-mentioned geometry feature in advance leads to
Disaggregated model output is crossed to the recognition result of component, to obtain primary segmentation as a result, being then based on visual signature to preliminary point
It cuts result to be modified, with generating unit segmentation result.Wherein, disaggregated model can pass through certain amount sample training.Sample
Originally it may include lossless component, also may include the component threedimensional model after being damaged in various accidents.Each component sample
Including component names or class label.Disaggregated model may include one or more models.It include multiple models in disaggregated model
In the case where, a disaggregated model can be trained for each component, for example, two disaggregated models of wheel training are directed to,
Component can be divided into wheel and non-wheel.In the case where disaggregated model includes a model, which can be more
The combination of a two disaggregated model, is also possible to disaggregated model more than one, can once identify all parts of damaged vehicle.Classification
Model can be the model of such as decision tree (Decision Tree), support vector machines (SVM) etc, and details are not described herein.
In one implementation, when view-based access control model feature is modified primary segmentation result, can by visual signature and just
Segmentation result Introduced Malaria model is walked, the component segmentation result of damaged vehicle is determined according to the output result of correction model.Wherein,
Correction model can be by the primary segmentation result and visual signature of sample, and the component segmentation result manually marked is instructed
Practice.By the model that primary segmentation result and visual signature input are selected, model is adjusted by the component segmentation result manually marked
Parameter obtains correction model.
In another realization, visual signature can also be mapped to threedimensional model, further according to preset modification rule, knot
Mapping result adjustment primary segmentation is closed as a result, to obtain component segmentation result.Wherein, visual signature is mapped to threedimensional model
Effect is as shown in Figure 4.Fig. 4 be a vehicle threedimensional model in be mapped with the effect picture after visual signature.By vision spy
During sign is mapped to threedimensional model, characteristic point on threedimensional model can be selected as datum mark, by visual signature according to
Datum mark carries out changes in coordinates, to combine with threedimensional model.It can recorde different components in modification rule and divide mistake
Accidentally and corresponding modification is tactful.For example, the component newly increased, can not be identified by classifier, corresponding modification in modification rule
Strategy can be the component that will be newly increased and split.Metal decorative strip, plaque on such as car door.For another example, two of the same side
Car door, since its apex coordinate position is adjacent, curved surface normal direction is similar, and usually, curved surface smoothness is close to smooth transition, therefore
It is difficult to preferably be split two car doors using only aforementioned geometries feature.In modification rule, for primary segmentation
As a result the corresponding modification strategy of vehicle door edge mistake in can be, and need to carry out accurate adjustment edge etc. according to visual signature.
According to another possible design, visual signature and geometry feature can also be inputted together to training in advance
Parted pattern determines the component segmentation result of damaged vehicle according to the output result of parted pattern.Wherein, parted pattern can be
Based on certain amount sample training.Sample may include the picture of original vehicle or the vehicle in various accidents after damage
And threedimensional model.Each sample has the component segmentation result manually marked.Geometry knot is extracted from picture and threedimensional model
The model that structure feature and visual signature input are selected, adjusts model parameter according to the component segmentation result manually marked, to instruct
Get parted pattern.
Above procedure is looked back, in the component cutting procedure to damaged vehicle, while obtaining the photo site of damaged vehicle
And threedimensional model, using the Image Visual Feature and three-dimensional structural feature of damaged vehicle, to determine the component segmentation of damaged vehicle
As a result.Since richer information of vehicles is utilized, segmentation result can be made more acurrate, to improve having for vehicle damage identification
Effect property.
According to the embodiment of another aspect, a kind of component segmenting device in vehicle damage identification process is also provided.
Fig. 5 shows the schematic block diagram for the component segmenting device in vehicle damage identification process according to one embodiment.Such as Fig. 5
It is shown, include: acquiring unit 51 for the component segmenting device 500 in vehicle damage identification process, is configured to obtain impaired vehicle
At least one photo site and threedimensional model；First extraction unit 52 is configured to extract damaged vehicle from threedimensional model
Geometry feature；Second extraction unit 53, the vision for being configured to extract damaged vehicle from least one photo site are special
Sign；Cutting unit 54 is configured at least view-based access control model feature, geometry feature and carries out component segmentation to damaged vehicle.
In this specification embodiment, at least one photo site of the available damaged vehicle of acquiring unit 51, and
Threedimensional model.Acquiring unit 51 can obtain damaged vehicle by the collection in worksite equipment of such as smart phone, camera etc
Photo site.Acquiring unit 51 can be a part of collection in worksite equipment, can also with collection in worksite equipment by wired or
Wireless network connection.
First extraction unit 52 can extract the geometry of damaged vehicle from threedimensional model acquired in acquiring unit 51
Feature, such as shape feature, volume characteristic, apex coordinate, curved surface normal angle, curved surface smoothness, unblind distance are retouched
It states one or more in function etc..
Meanwhile second extraction unit 53 can be extracted from least one photo site acquired in acquiring unit 51 it is impaired
The visual signature of vehicle.Such as it is one or more in the coherent feature in region, contour feature, colour consistency feature etc..
Cutting unit 54 can be mentioned at least based on the visual signature extracted by the first extraction unit 52, and by second
The geometry feature for taking unit 53 to extract carries out component segmentation to damaged vehicle.
According to a possible design, geometry feature can first be inputted classification mould trained in advance by cutting unit 54
Type, by disaggregated model output to the recognition result of component, to obtain primary segmentation as a result, being then based on visual signature to first
Step segmentation result is modified, with generating unit segmentation result.In one implementation, cutting unit 54 can by visual signature and
Primary segmentation result Introduced Malaria model, the component segmentation result of damaged parts is determined according to the output result of correction model.?
During another is realized, visual signature can also be mapped to threedimensional model by cutting unit 54, then according to preset modification rule,
In conjunction with mapping result adjustment primary segmentation as a result, to obtain the component segmentation result of damaged vehicle.Wherein, modification rule is for retouching
State the corresponding relationship of component partition error item and modification strategy.
According to another possible design, cutting unit 54 can also input together visual signature and geometry feature
Trained parted pattern in advance, the component segmentation result of damaged vehicle is determined according to the output result of parted pattern.
It is worth noting that device 500 shown in fig. 5 be with Fig. 2 shows the corresponding device of embodiment of the method implement
Example, Fig. 2 shows embodiment of the method in it is corresponding describe be equally applicable to device 500, details are not described herein.
By apparatus above, in the component cutting procedure to damaged vehicle, while the photo site of damaged vehicle is obtained
And threedimensional model, the two-dimensional Image Visual Feature of damaged vehicle and three-dimensional structural feature are made full use of, can be made to damaged vehicle
Component segmentation result it is more accurate, thus improve vehicle damage identification validity.
According to the embodiment of another aspect, a kind of computer readable storage medium is also provided, is stored thereon with computer journey
Sequence enables computer execute method described in conjunction with Figure 2 when the computer program executes in a computer.
According to the embodiment of another further aspect, a kind of calculating equipment, including memory and processor, the memory are also provided
In be stored with executable code, when the processor executes the executable code, realize the method in conjunction with described in Fig. 2.
Those skilled in the art are it will be appreciated that in said one or multiple examples, function described in the invention
It can be realized with hardware, software, firmware or their any combination.It when implemented in software, can be by these functions
Storage in computer-readable medium or as on computer-readable medium one or more instructions or code transmitted.
Above-described specific embodiment has carried out further the purpose of the present invention, technical scheme and beneficial effects
It is described in detail, it should be understood that being not intended to limit the present invention the foregoing is merely a specific embodiment of the invention
Protection scope, all any modification, equivalent substitution, improvement and etc. on the basis of technical solution of the present invention, done should all
Including within protection scope of the present invention.
Claims (16)
1. a kind of component dividing method for vehicle damage identification, which comprises
Obtain at least one photo site and threedimensional model of damaged vehicle；
The geometry feature of the damaged vehicle is extracted from the threedimensional model；
The visual signature of the damaged vehicle is extracted from least one described photo site；
Component segmentation is at least carried out to the damaged vehicle based on the visual signature, the geometry feature.
2. described to be at least based on the visual signature, the geometry feature according to the method described in claim 1, wherein
Carrying out component segmentation to the damaged vehicle includes:
By geometry feature input disaggregated model trained in advance, the identification to component is exported by the disaggregated model
As a result, to obtain primary segmentation result；
The primary segmentation result is modified based on the visual signature, with generating unit segmentation result.
3. described to be carried out based on the visual signature to the primary segmentation result according to the method described in claim 2, wherein
It corrects, includes: with generating unit segmentation result
It is true according to the output result of the correction model by the visual signature and the primary segmentation result Introduced Malaria model
The fixed component segmentation result.
4. described to be carried out based on the visual signature to the primary segmentation result according to the method described in claim 2, wherein
It corrects, includes: with generating unit segmentation result
The visual signature is mapped to the threedimensional model；
According to preset modification rule, the primary segmentation is adjusted in conjunction with mapping result as a result, to obtain the component segmentation knot
Fruit, wherein the modification rule is used to describe the corresponding relationship of component partition error item and modification strategy.
5. described to be at least based on the visual signature, the geometry feature according to the method described in claim 1, wherein
Carrying out component segmentation to the damaged vehicle includes:
By the visual signature, the geometry feature input parted pattern trained in advance, according to the parted pattern
Output result determines the component segmentation result.
6. according to the method described in claim 1, wherein, the geometry feature include at least one of the following: shape feature,
Volume characteristic, apex coordinate, curved surface normal angle, curved surface smoothness, unblind are apart from described function.
7. according to the method described in claim 1, wherein, the visual signature include at least one of the following: region link up feature,
Contour feature, colour consistency feature.
8. a kind of component segmenting device for vehicle damage identification, described device include:
Acquiring unit is configured to obtain at least one photo site and threedimensional model of damaged vehicle；
First extraction unit is configured to extract the geometry feature of the damaged vehicle from the threedimensional model；
Second extraction unit is configured to extract the visual signature of the damaged vehicle from least one described photo site；
Cutting unit is configured at least based on the visual signature, the geometry feature to damaged vehicle carry out portion
Part segmentation.
9. device according to claim 8, wherein the cutting unit is further configured to:
By geometry feature input disaggregated model trained in advance, the identification to component is exported by the disaggregated model
As a result, to obtain primary segmentation result；
The primary segmentation result is modified based on the visual signature, with generating unit segmentation result.
10. device according to claim 9, wherein the cutting unit is additionally configured to:
It is true according to the output result of the correction model by the visual signature and the primary segmentation result Introduced Malaria model
The fixed component segmentation result.
11. device according to claim 9, wherein the cutting unit is additionally configured to:
The visual signature is mapped to the threedimensional model；
According to preset modification rule, the primary segmentation is adjusted in conjunction with mapping result as a result, to obtain the component segmentation knot
Fruit, wherein the modification rule is used to describe the corresponding relationship of component partition error item and modification strategy.
12. device according to claim 8, wherein the cutting unit is further configured to:
By the visual signature, the geometry feature input parted pattern trained in advance, according to the parted pattern
Output result determines the component segmentation result.
13. device according to claim 8, wherein the geometry feature includes at least one of the following: shape spy
Sign, volume characteristic, apex coordinate, curved surface normal angle, curved surface smoothness, unblind are apart from described function.
14. device according to claim 8, wherein the visual signature includes at least one of the following: the coherent spy in region
Sign, contour feature, colour consistency feature.
15. a kind of computer readable storage medium, is stored thereon with computer program, when the computer program in a computer
When execution, computer perform claim is enabled to require the method for any one of 1-7.
16. a kind of calculating equipment, including memory and processor, which is characterized in that be stored with executable generation in the memory
Code realizes method of any of claims 1-7 when the processor executes the executable code.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811012740.1A CN109117847A (en) | 2018-08-31 | 2018-08-31 | Component dividing method and device for vehicle damage identification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811012740.1A CN109117847A (en) | 2018-08-31 | 2018-08-31 | Component dividing method and device for vehicle damage identification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109117847A true CN109117847A (en) | 2019-01-01 |
Family
ID=64860689
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811012740.1A Pending CN109117847A (en) | 2018-08-31 | 2018-08-31 | Component dividing method and device for vehicle damage identification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109117847A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110428403A (en) * | 2019-07-22 | 2019-11-08 | 宝能汽车有限公司 | Vehicle inspection method and electronic device |
| CN113706315A (en) * | 2021-07-28 | 2021-11-26 | 中国银行保险信息技术管理有限公司 | Vehicle damage assessment method, device and equipment |
| CN114418998A (en) * | 2022-01-20 | 2022-04-29 | 金瓜子科技发展（北京）有限公司 | Vehicle part defect detection method and system and electronic equipment |
| CN114913330A (en) * | 2022-07-18 | 2022-08-16 | 中科视语(北京)科技有限公司 | Point cloud component segmentation method and device, electronic equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103310449A (en) * | 2013-06-13 | 2013-09-18 | 沈阳航空航天大学 | Lung segmentation method based on improved shape model |
| CN106815847A (en) * | 2017-01-12 | 2017-06-09 | 非凡智慧（宁夏）科技有限公司 | Trees dividing method and single tree extracting method based on laser radar point cloud |
| CN107492107A (en) * | 2017-08-10 | 2017-12-19 | 昆山伟宇慧创智能科技有限公司 | The object identification merged based on plane with spatial information and method for reconstructing |
| CN107730503A (en) * | 2017-09-12 | 2018-02-23 | 北京航空航天大学 | The image object component-level semantic segmentation method and apparatus of three-dimensional feature insertion |
| US20180225539A1 (en) * | 2016-01-29 | 2018-08-09 | Pointivo, Inc. | Systems and methods for extracting information about objects from scene information |
-
2018
- 2018-08-31 CN CN201811012740.1A patent/CN109117847A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103310449A (en) * | 2013-06-13 | 2013-09-18 | 沈阳航空航天大学 | Lung segmentation method based on improved shape model |
| US20180225539A1 (en) * | 2016-01-29 | 2018-08-09 | Pointivo, Inc. | Systems and methods for extracting information about objects from scene information |
| CN106815847A (en) * | 2017-01-12 | 2017-06-09 | 非凡智慧（宁夏）科技有限公司 | Trees dividing method and single tree extracting method based on laser radar point cloud |
| CN107492107A (en) * | 2017-08-10 | 2017-12-19 | 昆山伟宇慧创智能科技有限公司 | The object identification merged based on plane with spatial information and method for reconstructing |
| CN107730503A (en) * | 2017-09-12 | 2018-02-23 | 北京航空航天大学 | The image object component-level semantic segmentation method and apparatus of three-dimensional feature insertion |
Non-Patent Citations (1)
| Title |
|---|
| 皮志明: ""结合深度信息的图像分割算法研究"", 《中国博士学位论文全文数据库 信息科技辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110428403A (en) * | 2019-07-22 | 2019-11-08 | 宝能汽车有限公司 | Vehicle inspection method and electronic device |
| CN113706315A (en) * | 2021-07-28 | 2021-11-26 | 中国银行保险信息技术管理有限公司 | Vehicle damage assessment method, device and equipment |
| CN114418998A (en) * | 2022-01-20 | 2022-04-29 | 金瓜子科技发展（北京）有限公司 | Vehicle part defect detection method and system and electronic equipment |
| CN114913330A (en) * | 2022-07-18 | 2022-08-16 | 中科视语(北京)科技有限公司 | Point cloud component segmentation method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109584294B (en) | Pavement point cloud extraction method and device based on laser point cloud | |
| CN109117847A (en) | Component dividing method and device for vehicle damage identification | |
| Lee et al. | Perceptual organization of 3D surface points | |
| CN106776996A (en) | Method and apparatus for testing the accuracy of high accuracy map | |
| JP2022522385A (en) | Road sign recognition methods, map generation methods, and related products | |
| CN103646416A (en) | Three-dimensional cartoon face texture generation method and device | |
| CN112613378A (en) | 3D target detection method, system, medium and terminal | |
| CN116433860B (en) | Intersection surface generation method, device, equipment, storage medium and program product | |
| US20220230453A1 (en) | Method and apparatus for generating zebra crossing in high resolution map, and electronic device | |
| CN105550688A (en) | Method and device for point cloud data classification | |
| JP3078166B2 (en) | Object recognition method | |
| Liu et al. | Image-translation-based road marking extraction from mobile laser point clouds | |
| CN112435336B (en) | Curve type identification method and device, electronic equipment and storage medium | |
| CN116468870B (en) | Three-dimensional visual modeling method and system for urban road | |
| CN102037327A (en) | Method of and computer implemented system for generating a junction view image | |
| CN117036634A (en) | Automatic construction method for three-dimensional scene of smart city | |
| CN104573722A (en) | Three-dimensional face race classifying device and method based on three-dimensional point cloud | |
| CN113487741A (en) | Dense three-dimensional map updating method and device | |
| CN118898838B (en) | Method, device, medium and vehicle for determining three-dimensional shape information of road obstacles | |
| CN115661444A (en) | Image processing method, device, equipment, storage medium and product | |
| JP7204087B2 (en) | Object recognition device | |
| CN112507891B (en) | Method and device for automatically identifying high-speed intersection and constructing intersection vector | |
| Omidalizarandi et al. | Segmentation and classification of point clouds from dense aerial image matching | |
| Nguyen et al. | High-Definition Texture Reconstruction for 3D Image-based Modelling. | |
| CN112528892A (en) | Unmanned aerial vehicle point cloud lane line extraction method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 40002329 Country of ref document: HK |
|
| TA01 | Transfer of patent application right |
Effective date of registration: 20200927 Address after: Cayman Enterprise Centre, 27 Hospital Road, George Town, Grand Cayman Islands Applicant after: Advanced innovation technology Co.,Ltd. Address before: A four-storey 847 mailbox in Grand Cayman Capital Building, British Cayman Islands Applicant before: Alibaba Group Holding Ltd. Effective date of registration: 20200927 Address after: Cayman Enterprise Centre, 27 Hospital Road, George Town, Grand Cayman Islands Applicant after: Innovative advanced technology Co.,Ltd. Address before: Cayman Enterprise Centre, 27 Hospital Road, George Town, Grand Cayman Islands Applicant before: Advanced innovation technology Co.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190101 |
|
| RJ01 | Rejection of invention patent application after publication |