CN118603111A - Multi-source sensor information fusion and verification method, device and computing equipment for road sweeper - Google Patents

Abstract

The present invention discloses a method, device and computing equipment for multi-source sensor information fusion and verification of a sweeper, wherein the method includes: obtaining data information of each sensor and preprocessing; dividing the data information of each sensor by characteristics to obtain an approximate linear data set and a non-Gaussian data set; for the approximate linear data set, estimating the prior probability distribution of the system state by an extended Kalman filter method; for the non-Gaussian data set, generating random particles by a particle filter method to approximate the posterior probability distribution of the system; dynamically adjusting the fusion weights of the extended Kalman filter method and the particle filter method to obtain the optimal fused environmental perception information; transmitting the verified environmental perception information to the control system of the sweeper for navigation, obstacle avoidance and path planning. The present invention improves the reliability of information fusion and verification by estimating data sets with different characteristics using an extended Kalman filter or a particle filter method.

Description

Multi-source sensor information fusion and verification method, device and computing equipment for road sweeper

Technical Field

The present invention relates to the technical field of intelligent transportation and sensor information fusion, and in particular to a method, device and computing equipment for fusion and verification of multi-source sensor information of a road sweeper.

Background Art

With the in-depth development of sweeper technology, its intelligence level is getting higher and higher. In order to achieve accurate navigation, obstacle avoidance and path planning of sweepers, it is necessary to accurately obtain and fuse information from multiple sensors. However, due to the complex and changeable working environment of sweepers, sensor data often has different characteristics, such as linear and nonlinear, Gaussian and non-Gaussian distribution, etc. How to effectively fuse sensor data becomes a key issue. However, before fusing sensor information, how to evaluate sensor information directly affects or determines the effect of information fusion.

Evaluating sensor information can identify and exclude data with low reliability, ensuring that the fusion process is based on high-quality data. In addition, it ensures that stable performance can be maintained in different environments and conditions, and that efficient and accurate perception and response capabilities can be maintained in complex and changing environments. Therefore, evaluating sensor information before fusion is a key step in ensuring the accuracy of the information fusion process, which is of great significance for improving the safety and reliability of the sweeper.

Existing sensor information evaluation methods include threshold checking method, statistical analysis method, redundant sensor comparison method and neural network model. Among them, the threshold checking method relies on experience or experiments. Different environments or working conditions require different thresholds and cannot adapt to dynamically changing environments. The statistical analysis method performs statistical analysis on sensor data, such as calculating statistical quantities such as mean, standard deviation, skewness, kurtosis, etc., and cannot capture the complex and changing distribution changes of multiple data. The redundant sensor comparison method uses multiple sensors of the same or different types to measure the same physical quantity. If the results differ too much, it means that there is a problem with one of the sensors, which requires additional hardware costs. The neural network model requires a large amount of labeled data to train the learning model, and has poor robustness and generalization.

In the process of researching and developing solutions to the problem of how to fuse and verify multi-source sensor information, the inventors of this application found that the methods for fusing and verifying multi-source sensor information are relatively mature and difficult to improve. However, how to evaluate the uncertainty of current sensor information will dynamically determine the fusion weight (weight distribution). Therefore, the inventors took a different approach and converted the research on fusing and verifying multi-source sensor information into research on evaluating sensor information.

Summary of the invention

In view of the above problems, the present invention is proposed to provide a method, device and computing equipment for multi-source sensor information fusion and verification of a sweeper to improve the reliability of multi-source sensor information fusion and verification. The cover device can be applied to sweepers of various power types, especially hybrid power sweepers. The specific technical solution is as follows:

According to one aspect of the present invention, a method for fusion and verification of multi-source sensor information of a road sweeper is provided, comprising:

At least one camera is disposed on the top, front and both sides of the sweeper, a plurality of laser radars are disposed around, at least one ultrasonic sensor is disposed on the front and rear bumpers, and at least one inertial sensor is disposed on the chassis;

Acquire data information from each sensor and perform preprocessing, wherein the preprocessing includes data denoising and outlier removal;

The data information of each sensor is divided into characteristics to obtain an approximate linear data set and a non-Gaussian data set; for the approximate linear data set, the prior probability distribution of the system state is estimated by the extended Kalman filter method, and the state estimation is corrected in combination with the latest observation data; for the non-Gaussian data set, random particles are generated by the particle filter method to approximately estimate the posterior probability distribution of the system, and the weight of each particle is calculated according to the historical observation data, and the state estimation is obtained by weighted average;

The fusion weights of the extended Kalman filter method and the particle filter method are dynamically adjusted through the adaptive fusion method to obtain the optimal fusion environment perception information;

The fused environmental perception information is subjected to consistency check and redundancy check, and the checked environmental perception information is transmitted to the control system of the sweeper for use in navigation, obstacle avoidance and path planning.

In an optional manner, the data denoising method further comprises:

Preset the filter window size and the P-order polynomial order according to the raw data sequence of the multiple sensor data information;

For each data point in the original data series ( ), select A local data window centered on the Nth dimension and containing N data points is used; a P-order polynomial is fitted on the local data window using the least squares method to obtain fitting polynomial coefficients;

Use this fitting polynomial in Evaluate at and get the filtered data point ( ),Will The new filtered data sequence is combined in sequence; the filtered data sequence is used as the sensor data information after high-frequency noise is removed.

In an optional manner, the method for removing outliers further comprises:

According to the raw data sequences of multiple sensor data points, the mean value (μ) and standard deviation (σ) of all data points in the raw data sequences are calculated;

For each data point in the original data series ( ), if | -μ| ≤ 3σ, then the data point is a normal value and is retained in the data series; if | - μ| > 3σ, then the data point is an outlier and is removed from the data sequence;

The processed data sequence is used as the sensor data information after removing outliers.

In an optional manner, the characteristic division of the data information of each sensor to obtain an approximately linear data set and a non-Gaussian data set further includes:

Calculate the P value of the data information of each sensor by the Anderson-Darling test method, and classify the data information with the P value greater than or equal to the preset significance level as a non-Gaussian data set;

For data information with a P value less than the preset significance level, the R-square value is obtained by fitting the linear regression model, and the data information with an R-square value greater than or equal to the preset value is divided into an approximate linear data set.

In an optional manner, estimating the prior probability distribution of the system state by using the extended Kalman filter method and correcting the state estimation in combination with the latest observation data further includes:

Step S1, set the initial state estimate ( ) and the initial covariance matrix ( );

Step S2, according to the nonlinear state transfer function ( ) Predict the current state estimate ( );

Step S3, calculate the state transfer function ( )’s Jacobian matrix ( ), according to the Jacobian matrix ( ) and the process noise covariance matrix ( ) Prediction covariance( );

Step S4, according to the observation function ( )’s Jacobian matrix ( ) and the observation noise covariance matrix ( ) Calculate the Kalman gain ( );

Step S5, according to the observed value ( ) and Kalman gain ( ) Update state estimate( );

Step S6, according to the Kalman gain ( ), Jacobian matrix( ) and covariance ( ) Update the covariance matrix ( );

Step S7, estimate the current state ( ) and the covariance matrix ( ) as the initial value for the next moment, and repeat steps S2 and S3.

In an optional manner, in step S2, the current state estimation ( ) is calculated as:

in, is the state estimate at the previous moment, is the control input at the current moment;

In step S3, the covariance ( ) is calculated as:

in, is the covariance matrix of the state estimate at the previous moment obtained after updating the previous time step in the extended Kalman filter context;

In step S4, the Kalman gain ( ) is calculated as:

In step S5, the state estimation ( ) is calculated as:

In step S6, the covariance matrix ( ) is calculated as:

Where I is the identity matrix.

In an optional manner, the generating random particles by a particle filtering method to approximate the posterior probability distribution of the system further comprises:

Step S21, at time step (t=0), from the prior distribution ( ) and extract N particles ( ), assigning equal weight to each particle;

Step S22, extract a sample for each particle from the importance distribution ( );

Step S23, calculate the weight of each particle according to the observation likelihood and state transition probability ( );

Step S24, estimating the posterior probability distribution of the system based on the particles and their weights.

In an optional manner, in step S23, the weight of each particle is:

in, is the observation likelihood, indicating that given the state Observed The probability of is the state transition probability, which is from state Transfer to The state transition probability, N is the number of particles;

In step S24, the calculation formula of the posterior probability distribution is:

in, is the Dirac function, which represents the particle The position in state space, represents the state of the ith particle at time step t, is the normalized weight of the ith particle at time step t, Indicates that given historical data Under the condition of The probability distribution of is the data sequence observed by the sensor.

According to another aspect of the present invention, a multi-source sensor information fusion and verification device for a sweeper is provided, comprising:

A setting module is used to set at least one camera on the top, front and both sides of the sweeper, set multiple laser radars around, set at least one ultrasonic sensor on the front and rear bumpers, and set at least one inertial sensor on the chassis;

A preprocessing module, used to obtain data information from each sensor and perform preprocessing, wherein the preprocessing includes data denoising and outlier removal;

A state estimation module is used to divide the data information of each sensor into characteristics to obtain an approximate linear data set and a non-Gaussian data set; for the approximate linear data set, the prior probability distribution of the system state is estimated by the extended Kalman filter method, and the state estimation is corrected in combination with the latest observation data; for the non-Gaussian data set, random particles are generated by the particle filter method to approximately estimate the posterior probability distribution of the system, and the weight of each particle is calculated according to the historical observation data, and the state estimation is obtained by weighted average;

A fusion module is used to dynamically adjust the fusion weights of the extended Kalman filter method and the particle filter method through an adaptive fusion method to obtain the optimal fusion environment perception information;

The verification module is used to perform consistency check and redundancy check on the fused environmental perception information, and transmit the verified environmental perception information to the control system of the sweeper for navigation, obstacle avoidance and path planning.

According to another aspect of the present invention, there is provided a computing device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned multi-source sensor information fusion and verification method of the sweeper.

According to the solution provided by the present invention, at least one camera is arranged on the top, front and both sides of the sweeper, multiple laser radars are arranged around, at least one ultrasonic sensor is arranged on the front and rear bumpers, and at least one inertial sensor is arranged on the chassis; data information of each sensor is obtained and preprocessed, wherein the preprocessing includes data denoising and outlier removal; the data information of each sensor is characterized to obtain an approximate linear data set and a non-Gaussian data set; for the approximate linear data set, the prior probability distribution of the system state is estimated by the extended Kalman filter method, and the state estimation is corrected in combination with the latest observation data; for the non-Gaussian data set, random particles are generated by the particle filter method to approximate the posterior probability distribution of the system, and the weight of each particle is calculated according to the historical observation data, and the state estimation is obtained by weighted average; the fusion weights of the extended Kalman filter method and the particle filter method are dynamically adjusted by the adaptive fusion method to obtain the optimal fusion environment perception information; the fusion environment perception information is checked for consistency and redundancy, and the verified environment perception information is transmitted to the control system of the sweeper for navigation, obstacle avoidance and path planning. The present invention converts the research on fusion and verification of multi-source sensor information into the research on evaluation of sensor information, and improves the reliability of information fusion and verification by estimating data sets with different characteristics using extended Kalman filtering or particle filtering methods. Specifically, by setting cameras, laser radars, ultrasonic sensors and inertial sensors on multiple key parts such as the top, front, both sides and chassis of the sweeper, all-round perception of the surrounding environment is achieved, which not only ensures the comprehensiveness of information, but also improves the accuracy of information through the complementarity between different sensors. Data denoising uses the least squares fitting P-order polynomial method in the local data window to remove high-frequency noise and retain useful information. Outlier removal calculates the mean and standard deviation of data points, regards data points exceeding a certain threshold as outliers and removes them, further reducing the impact of noise and interference on information fusion. The extended Kalman filter accurately estimates the state of the system through prediction and update steps, and the particle filter approximates the posterior probability distribution of the system by generating random particles and calculating their weights. By estimating data sets with different characteristics using extended Kalman filtering or particle filtering methods, the reliability of information fusion and verification is further improved. By dynamically adjusting the fusion weights of the extended Kalman filter and the particle filter through an adaptive fusion method, the sweeper can intelligently select the fusion strategy that best suits the current environmental conditions, ensuring the accuracy and real-time nature of environmental perception information, and achieving optimal performance in both linear and nonlinear, Gaussian and non-Gaussian environments. In addition, consistency checks and redundancy checks are performed on the fused environmental perception information, further enhancing the reliability and robustness of the information and reducing the risk of misjudgment and misoperation. As a result, the sweeper is more efficient and safer in navigation, obstacle avoidance, and path planning.

The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below.

BRIEF DESCRIPTION OF THE DRAWINGS

Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

FIG1 is a schematic diagram showing a flow chart of a method for fusion and verification of multi-source sensor information of a road sweeper according to an embodiment of the present invention;

FIG2 shows a schematic structural diagram of a multi-source sensor information fusion and verification device for a road sweeper according to an embodiment of the present invention;

FIG. 3 shows a schematic diagram of the structure of a computing device according to an embodiment of the present invention.

DETAILED DESCRIPTION

The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to enable the scope of the present invention to be fully communicated to those skilled in the art.

How to further improve the system's adaptability to uncertainty in dynamic environments based on the existing relatively mature information fusion and verification methods. After in-depth research, the inventors found that although simply optimizing the fusion and verification algorithm is important, the more critical thing is how to accurately evaluate the uncertainty of the current sensor information and dynamically determine the fusion weight based on this uncertainty. Therefore, a new approach was taken to shift the research focus from the traditional multi-source sensor information fusion and verification methods to how to evaluate the uncertainty of sensor information. By using extended Kalman filtering or particle filtering methods to estimate data sets with different characteristics, the information uncertainty of different sensors under different environmental conditions can be accurately evaluated, which improves the reliability of information fusion and verification, thereby improving the accuracy and reliability of the overall perception information.

FIG1 shows a schematic flow chart of a method for fusion and verification of multi-source sensor information of a road sweeper according to an embodiment of the present invention. Specifically, as shown in FIG1 , the method includes the following steps:

Step S101, at least one camera is arranged on the top, front and both sides of the sweeper, multiple laser radars are arranged around, at least one ultrasonic sensor is arranged on the front and rear bumpers, and at least one inertial sensor is arranged on the chassis.

The camera provides high-definition visual information around the sweeper, capturing detailed image data from the top, front, and sides. LiDAR accurately measures the distance between the sweeper and surrounding objects by emitting laser beams and receiving reflected signals, providing high-precision three-dimensional environmental information. Ultrasonic sensors are used for close-range obstacle detection, especially for complex or low-light environments. Inertial sensors monitor the movement of the sweeper in real time, providing accurate data support for navigation and path planning, ensuring that the sweeper can operate along the optimal path. Comprehensive environmental perception can ensure the safety of cleaning operations. The specific installation locations of the above sensors are shown in Table 1:

Table 1

Installation location Sensor Type Reference location Reference Model Top center Camera About 20cm from the edge of the roof xx-2CD2T85G4-I Top center (30cm from camera head) LiDAR Next to the camera xx-LiDAR-M1 Front center Camera Angle to the front windshield xx-2CD2T85G4-I Front bumper top LiDAR About 80cm from the ground xx-LiDAR-M1 Front bumper left and right sides Ultrasonic Sensors About 50cm from the ground HC-SR04 Both sides of the vehicle Camera Angle with the car body xx-2CD2T85G4-I Both sides of the vehicle LiDAR Determined according to the size and structure of the sweeper xx-LiDAR-M1 Four corners or sides of the vehicle LiDAR Determined by the size of the sweeper and the radar scanning range xx-LiDAR-M1 Rear bumper left and right sides Ultrasonic Sensors About 50cm from the ground xx-SR04 Chassis center Inertial Sensors Central location of suspension xx IS16488

Step S102, acquiring data information of each sensor and performing preprocessing, wherein the preprocessing includes data denoising and outlier removal.

The data information of each sensor is shown in Table 2:

Table 2

Sensor Type Data Information Camera Image data: JPEG/PNG format, including images of the environment around the sweeper Video data: MP4 format, 30 frames per second, including real-time video of the sweeper in motion LiDAR Point cloud data: contains three-dimensional coordinate points, each point contains (x, y, z) coordinates and reflection intensity, such as: [(1.2, 0.5,0.8, 120), (1.3, 0.6, 0.7, 100), ...] Ultrasonic Sensors Distance value: Real-time measurement of the distance between the sweeper and the object in front, such as: 2.5m, 1.8m, 3.2m, ... Inertial Sensors Acceleration data: three-axis acceleration value (x, y, z), unit is g; Angular velocity data: three-axis angular velocity value (x, y, z), unit is degree/second; such as: acceleration (0.1, -0.2, 0.05), angular velocity (0.01, 0.02, -0.01)

Data denoising processing:

For image or video data, you can use image filtering techniques (such as Gaussian filtering and median filtering) to remove noise from the image. For dynamic images, you can also use time filtering (such as frame difference method) to reduce the impact of noise. For lidar data, you can use statistical filtering (such as voxel filtering) to reduce the number of points while maintaining shape features, and you can also use methods such as random sampling consistency (RANSAC) to remove outliers. For ultrasonic sensor data, you can reduce the impact of noise by taking the average of multiple measurements. For inertial sensor data, you can use a low-pass filter to remove the noise contained. For acceleration data, you can remove noise by integrating the velocity and displacement.

Outlier removal processing:

For image data, you can remove abnormal pixels by setting the pixel value range or color threshold, and you can also use algorithms such as background subtraction and foreground detection to identify and remove abnormal areas. For point cloud data, you can remove abnormal points by setting the distance threshold or reflection intensity threshold. For ultrasonic sensor data, you can set the distance threshold or change rate threshold to remove outliers. For inertial sensor data, you can use statistical methods to identify and remove outliers, and you can also combine other sensor data (such as GPS) to verify the accuracy of inertial sensor data and remove outliers.

In an optional implementation, the data denoising method further comprises:

Preset the filter window size and the P-order polynomial order according to the raw data sequence of the multiple sensor data information;

For each data point in the original data series ( ), select A local data window centered on the Nth dimension and containing N data points is used; a P-order polynomial is fitted on the local data window using the least squares method to obtain fitting polynomial coefficients;

Use this fitting polynomial in Evaluate at and get the filtered data point ( ),Will The new filtered data sequence is combined in sequence; the filtered data sequence is used as the sensor data information after high-frequency noise is removed.

The data denoising method in this embodiment is suitable for a variety of sensor data types, including images, videos, lidar point clouds, ultrasonic data, and inertial sensor data. By presetting the filter window size and polynomial order, customized denoising is performed on different types of data. The filter window size and polynomial order are adjusted according to the characteristics of the actual data and the denoising requirements to achieve the best denoising effect. For different types of sensor data, different denoising techniques (such as image filtering, time filtering, statistical filtering, etc.) are selected for preprocessing. The method of fitting polynomials using local data windows and least squares method is used to quickly and accurately estimate the true value of the data point and reduce the impact of noise. The data is smoothed by fitting polynomials to effectively remove high-frequency noise while retaining the overall trend and characteristics of the data. Verification and removal of outliers in combination with other sensor data (such as GPS) can further improve the accuracy and reliability of the data. The denoising method in this embodiment reduces the workload of writing different denoising codes for different sensor data and improves development efficiency.

For example, the original data sequence shown in Table 3:

Table 3

Original data point index Original data value 1 10.2 2 10.5 ... ... ... i-2 11.0 i-1 10.8 i 10.6 i+1 10.4 i+2 10.3 ... ... ...

For data points (10.6), select The local data window (N=5) with N data points as the center is arrive Data points (as shown in Table 4):

Table 4

Local Data Window Data Value 11.0 10.8 10.6 10.4 10.3

On this local data window, use the least squares method to fit a P-order polynomial (such as P=2) to obtain the coefficients of the fitting polynomial. Use the modified fitting polynomial to Evaluate at to get the filtered data point Finally, all the filtered data points The new filtered data sequence is combined in sequence as the sensor data information after removing high-frequency noise.

In an optional implementation, the method for removing outliers further comprises:

According to the raw data sequences of multiple sensor data points, the mean value (μ) and standard deviation (σ) of all data points in the raw data sequences are calculated;

For each data point in the original data series ( ), if | -μ| ≤ 3σ, then the data point is a normal value and is retained in the data series; if | - μ| > 3σ, then the data point is an outlier and is removed from the data sequence;

The processed data sequence is used as the sensor data information after removing outliers.

In this embodiment, the mean (μ) and standard deviation (σ) statistics are used as benchmarks to objectively measure the degree to which data points deviate from the normal range. The 3σ principle can capture approximately 99.7% of data points, while identifying and eliminating very few extreme values. It can quickly detect and eliminate outliers for a large amount of sensor data, ensuring the stability and reliability of the data stream. In addition, it does not depend on a specific sensor type or data type (including numerical type, image type, etc.). For different sensor data, you only need to calculate the corresponding mean and standard deviation, and then apply this method to eliminate outliers.

Step S103, the data information of each sensor is divided into characteristics to obtain an approximate linear data set and a non-Gaussian data set; for the approximate linear data set, the prior probability distribution of the system state is estimated by the extended Kalman filter method, and the state estimation is corrected in combination with the latest observation data; for the non-Gaussian data set, random particles are generated by the particle filter method to approximate the posterior probability distribution of the system, and the weight of each particle is calculated according to the historical observation data, and the state estimation is obtained by weighted average.

At present, ARIMA models or deep learning models are often used to estimate system states. However, ARIMA models have many parameters, are difficult to adjust, and require sufficient data for fitting and prediction. If the data length is insufficient, the model will be unstable or the prediction effect will be poor. The working principle and decision-making process inside the deep learning model are "black boxes", making the prediction results difficult to explain and trust.

In this embodiment, for approximate linear data sets, the extended Kalman filter method is used for iterative prediction and updating, which can more accurately estimate the prior probability distribution of the system state, and correct the state estimate in combination with the latest observation data, thereby improving the accuracy of data processing. For non-Gaussian data sets, random particles are generated by the particle filter method to approximate the posterior probability distribution of the system. Even when there is noise or error in the sensor data, the random particles generated by the particle filter method can offset these effects, thereby improving the accuracy of state estimation.

In an optional implementation, the characteristic division of the data information of each sensor to obtain an approximately linear data set and a non-Gaussian data set further includes:

Calculate the P value of the data information of each sensor by the Anderson-Darling test method, and classify the data information with the P value greater than or equal to the preset significance level as a non-Gaussian data set;

For data information with a P value less than the preset significance level, the R-square value is obtained by fitting the linear regression model, and the data information with an R-square value greater than or equal to the preset value is divided into an approximate linear data set.

In this embodiment, the Anderson-Darling test method is an extension of the Kolmogorov-Smirnov test method, which is particularly suitable for situations with small sample sizes. The Anderson-Darling test method can accurately determine whether the data obeys a Gaussian distribution, thereby accurately distinguishing non-Gaussian data sets. At the same time, the linearity of the data is evaluated by the R-square value of the linear regression model, which improves the recognition accuracy of the approximately linear data set.

In an optional implementation, estimating the prior probability distribution of the system state by using the extended Kalman filter method and correcting the state estimate in combination with the latest observation data further includes:

Step S1, set the initial state estimate ( ) and the initial covariance matrix ( );

Step S2, according to the nonlinear state transfer function ( ) Predict the current state estimate ( );

Step S3, calculate the state transfer function ( )’s Jacobian matrix ( ), according to the Jacobian matrix ( ) and the process noise covariance matrix ( ) Prediction covariance( );

Step S4, according to the observation function ( )’s Jacobian matrix ( ) and the observation noise covariance matrix ( ) Calculate the Kalman gain ( );

Step S5, according to the observed value ( ) and Kalman gain ( ) Update state estimate( );

Step S6, according to the Kalman gain ( ), Jacobian matrix( ) and covariance ( ) Update the covariance matrix ( );

Step S7, estimate the current state ( ) and the covariance matrix ( ) as the initial value for the next moment, and repeat steps S2 and S3.

In this embodiment, the extended Kalman filter can update the state estimate and covariance matrix at each time step, which is suitable for data processing scenarios with high real-time requirements. At the same time, due to its iterative update characteristics, the extended Kalman filter has high computational efficiency. The extended Kalman filter can fuse data from different sensors such as GPS, radar, and cameras to provide more accurate state estimation.

In this embodiment, in step S2, the current state estimation ( ) is calculated as:

in, is the state estimate at the previous moment, is the control input at the current moment;

In step S3, the covariance ( ) is calculated as:

in, is the covariance matrix of the state estimate at the previous moment obtained after updating the previous time step in the extended Kalman filter context;

In step S4, the Kalman gain ( ) is calculated as:

In step S5, the state estimation ( ) is calculated as:

In step S6, the covariance matrix ( ) is calculated as:

Where I is the identity matrix.

In this embodiment, the state estimate at the current moment can be directly calculated based on the state estimate at the previous moment and the control input at the current moment through the state transfer function, which can provide more accurate predictions for approximately linear systems. The Kalman gain comprehensively considers the uncertainty of the predicted state and the uncertainty of the observed data, and can adaptively balance the weights between the predicted value and the observed value, thereby improving the accuracy of the state estimate. The state estimate is corrected by the Kalman gain, and the observed information is fully utilized to correct the state estimate in real time, thereby improving the adaptability of the system. By reducing the covariance component corresponding to the utilized observed information, the uncertainty estimate of the unobserved information is improved.

In an optional implementation, the generating random particles by a particle filtering method to approximate the posterior probability distribution of the system further comprises:

Step S21, at time step (t=0), from the prior distribution ( ) and extract N particles ( ), assigning equal weight to each particle;

Step S22, extract a sample for each particle from the importance distribution ( );

Step S23, calculate the weight of each particle according to the observation likelihood and state transition probability ( );

Step S24, estimating the posterior probability distribution of the system based on the particles and their weights.

In this embodiment, by extracting samples from the importance distribution, the dynamic change process of the system is simulated, and various non-Gaussian and other complex distributions can be processed. The particle weight is calculated according to the observation likelihood and the state transition probability, and the observation information is fully utilized to evaluate the particles, thereby improving the accuracy of state estimation. Based on the particles and their weights, the posterior probability distribution of the system is estimated, and a non-parametric probability density estimation method is provided, which can process any form of probability distribution and has strong adaptability.

In an optional implementation, in step S23, the weight of each particle is:

in, is the observation likelihood, indicating that given the state Observed The probability of is the state transition probability, which is from state Transfer to The state transition probability, N is the number of particles;

In step S24, the calculation formula of the posterior probability distribution is:

in, is the Dirac function, which represents the particle The position in state space, represents the state of the ith particle at time step t, is the normalized weight of the ith particle at time step t, Indicates that given historical data Under the condition of The probability distribution of is the data sequence observed by the sensor.

In this embodiment, by giving the state Observed The probability of the state that best matches the current observation can be more accurately identified in an environment with noise and uncertainty. At the same time, considering the probability of transitioning from the previous state to the current state, the consistency and accuracy of particles can be maintained in continuous time steps to avoid over-reliance on observation data. Representing the posterior probability distribution of the system by particles and their weights helps engineers better understand the dynamic characteristics of the system and make decisions and controls accordingly. At the same time, it can quickly process a large number of particles and estimate the posterior probability distribution.

Step S104, dynamically adjusting the fusion weights of the extended Kalman filter method and the particle filter method through an adaptive fusion method to obtain optimal fusion environment perception information.

In this embodiment, the adaptive fusion method can dynamically adjust the fusion weights of EKF and PF according to the real-time environment to better adapt to various complex environments and improve the accuracy of perception information. For example, fuzzy sets and fuzzy rules are defined by fuzzy logic methods to describe the performance of EKF and PF and the relationship between them, the performance of EKF and PF is evaluated and the membership on the fuzzy set is determined, and the optimal fusion weight is calculated according to the fuzzy rules and membership.

Step S105, performing consistency check and redundancy verification on the fused environmental perception information, and transmitting the verified environmental perception information to the control system of the sweeper for use in navigation, obstacle avoidance and path planning.

For example, for consistency check, clarify the consistency standard of environmental perception information (such as data obtained by different sensors should match each other within a certain error range). Compare environmental perception data from different sensors to check whether the differences are within the consistency standard range. If the data comparison finds inconsistencies, take corrective measures (such as recalibrating sensors and adjusting sensor parameters). For redundancy verification, identify redundant data from different sensors or at different time points, and use redundant data to improve the accuracy of environmental perception information (such as integrating redundant data using weighted average or voting methods). Convert the verified environmental perception information into a format that the sweeper control system can understand and process, and use transmission protocols such as CAN bus to ensure that the transmission speed of environmental perception information meets the requirements of the sweeper control system. Finally, using the fused environmental perception information, the sweeper can achieve autonomous navigation (planning the path from the current position to the target position), obstacle avoidance (by detecting obstacles and predicting their movement trajectories, the sweeper can adjust the path in real time to avoid collisions with obstacles) and path planning (planning the optimal cleaning path based on environmental perception information and the requirements of the cleaning task).

The solution provided by the above embodiment of the present invention is that at least one camera is arranged on the top, front and both sides of the sweeper, multiple laser radars are arranged around, at least one ultrasonic sensor is arranged on the front and rear bumpers, and at least one inertial sensor is arranged on the chassis; data information of each sensor is obtained and preprocessed, wherein the preprocessing includes data denoising and outlier removal; the data information of each sensor is characterized to obtain an approximate linear data set and a non-Gaussian data set; for the approximate linear data set, the prior probability distribution of the system state is estimated by the extended Kalman filter method, and the state estimation is corrected in combination with the latest observation data; for the non-Gaussian data set, random particles are generated by the particle filter method, the posterior probability distribution of the system is approximately estimated, and the weight of each particle is calculated according to the historical observation data, and the state estimation is obtained by weighted average; the fusion weights of the extended Kalman filter method and the particle filter method are dynamically adjusted by the adaptive fusion method to obtain the optimal fusion environment perception information; the fusion environment perception information is checked for consistency and redundancy, and the verified environment perception information is transmitted to the control system of the sweeper for navigation, obstacle avoidance and path planning. The present invention converts the research on fusion and verification of multi-source sensor information into the research on evaluation of sensor information, and improves the reliability of information fusion and verification by estimating data sets with different characteristics using extended Kalman filtering or particle filtering methods. Specifically, by setting cameras, laser radars, ultrasonic sensors and inertial sensors on multiple key parts such as the top, front, both sides and chassis of the sweeper, all-round perception of the surrounding environment is achieved, which not only ensures the comprehensiveness of information, but also improves the accuracy of information through the complementarity between different sensors. Data denoising uses the least squares fitting P-order polynomial method in the local data window to remove high-frequency noise and retain useful information. Outlier removal calculates the mean and standard deviation of data points, regards data points exceeding a certain threshold as outliers and removes them, further reducing the impact of noise and interference on information fusion. The extended Kalman filter accurately estimates the state of the system through prediction and update steps, and the particle filter approximates the posterior probability distribution of the system by generating random particles and calculating their weights. By estimating data sets with different characteristics using extended Kalman filtering or particle filtering methods, the reliability of information fusion and verification is further improved. By dynamically adjusting the fusion weights of the extended Kalman filter and the particle filter through an adaptive fusion method, the sweeper can intelligently select the fusion strategy that best suits the current environmental conditions, ensuring the accuracy and real-time nature of environmental perception information, and achieving optimal performance in both linear and nonlinear, Gaussian and non-Gaussian environments. In addition, consistency checks and redundancy checks are performed on the fused environmental perception information, further enhancing the reliability and robustness of the information and reducing the risk of misjudgment and misoperation. As a result, the sweeper is more efficient and safer in navigation, obstacle avoidance, and path planning.

2 shows a schematic diagram of the structure of a multi-source sensor information fusion and verification device for a road sweeper according to an embodiment of the present invention. The multi-source sensor information fusion and verification device 200 for a road sweeper comprises: a setting module 210 , a preprocessing module 220 , a state estimation module 230 , a fusion module 240 and a verification module 250 .

The setting module 210 is used to set at least one camera on the top, front and both sides of the sweeper, set multiple laser radars around, set at least one ultrasonic sensor on the front and rear bumpers, and set at least one inertial sensor on the chassis;

The preprocessing module 220 is used to obtain data information of each sensor and perform preprocessing, wherein the preprocessing includes data denoising and outlier removal;

The state estimation module 230 is used to divide the data information of each sensor into characteristics to obtain an approximate linear data set and a non-Gaussian data set; for the approximate linear data set, the prior probability distribution of the system state is estimated by the extended Kalman filter method, and the state estimation is corrected in combination with the latest observation data; for the non-Gaussian data set, random particles are generated by the particle filter method to approximately estimate the posterior probability distribution of the system, and the weight of each particle is calculated according to the historical observation data, and the state estimation is obtained by weighted average;

The fusion module 240 is used to dynamically adjust the fusion weights of the extended Kalman filter method and the particle filter method through an adaptive fusion method to obtain optimal fusion environment perception information;

The verification module 250 is used to perform consistency check and redundancy check on the fused environmental perception information, and transmit the verified environmental perception information to the control system of the sweeper for navigation, obstacle avoidance and path planning.

In an optional manner, the data denoising method further comprises:

Preset the filter window size and the P-order polynomial order according to the raw data sequence of the multiple sensor data information;

For each data point in the original data series ( ), select A local data window centered on the Nth dimension and containing N data points is used; a P-order polynomial is fitted on the local data window using the least squares method to obtain fitting polynomial coefficients;

Use this fitting polynomial in Evaluate at and get the filtered data point ( ),Will The new filtered data sequence is combined in sequence; the filtered data sequence is used as the sensor data information after high-frequency noise is removed.

In an optional manner, the method for removing outliers further comprises:

According to the raw data sequences of multiple sensor data points, the mean value (μ) and standard deviation (σ) of all data points in the raw data sequences are calculated;

For each data point in the original data series ( ), if | -μ| ≤ 3σ, then the data point is a normal value and is retained in the data series; if | - μ| > 3σ, then the data point is an outlier and is removed from the data sequence;

The processed data sequence is used as the sensor data information after removing outliers.

In an optional manner, the characteristic division of the data information of each sensor to obtain an approximately linear data set and a non-Gaussian data set further includes:

Calculate the P value of the data information of each sensor by the Anderson-Darling test method, and classify the data information with the P value greater than or equal to the preset significance level as a non-Gaussian data set;

For data information with a P value less than the preset significance level, the R-square value is obtained by fitting the linear regression model, and the data information with an R-square value greater than or equal to the preset value is divided into an approximate linear data set.

In an optional manner, estimating the prior probability distribution of the system state by using the extended Kalman filter method and correcting the state estimation in combination with the latest observation data further includes:

Step S1, set the initial state estimate ( ) and the initial covariance matrix ( );

Step S2, according to the nonlinear state transfer function ( ) Predict the current state estimate ( );

Step S3, calculate the state transfer function ( )’s Jacobian matrix ( ), according to the Jacobian matrix ( ) and the process noise covariance matrix ( ) Prediction covariance( );

Step S4, according to the observation function ( )’s Jacobian matrix ( ) and the observation noise covariance matrix ( ) Calculate the Kalman gain ( );

Step S5, according to the observed value ( ) and Kalman gain ( ) Update state estimate( );

Step S6, according to the Kalman gain ( ), Jacobian matrix( ) and covariance ( ) Update the covariance matrix ( );

Step S7, estimate the current state ( ) and the covariance matrix ( ) as the initial value for the next moment, and repeat steps S2 and S3.

In an optional manner, in step S2, the current state estimation ( ) is calculated as:

in, is the state estimate at the previous moment, is the control input at the current moment;

In step S3, the covariance ( ) is calculated as:

in, is the covariance matrix of the state estimate at the previous moment obtained after updating the previous time step in the extended Kalman filter context;

In step S4, the Kalman gain ( ) is calculated as:

In step S5, the state estimation ( ) is calculated as:

In step S6, the covariance matrix ( ) is calculated as:

Where I is the identity matrix.

In an optional manner, the generating random particles by a particle filtering method to approximate the posterior probability distribution of the system further comprises:

Step S21, at time step (t=0), from the prior distribution ( ) and extract N particles ( ), assigning equal weight to each particle;

Step S22, extract a sample for each particle from the importance distribution ( );

Step S23, calculate the weight of each particle according to the observation likelihood and state transition probability ( );

Step S24, estimating the posterior probability distribution of the system based on the particles and their weights.

In an optional manner, in step S23, the weight of each particle is:

in, is the observation likelihood, indicating that given the state Observed The probability of is the state transition probability, which is from state Transfer to The state transition probability, N is the number of particles;

In step S24, the calculation formula of the posterior probability distribution is:

in, is the Dirac function, which represents the particle The position in state space, represents the state of the ith particle at time step t, is the normalized weight of the ith particle at time step t, Indicates that given historical data Under the condition of The probability distribution of is the data sequence observed by the sensor.

The solution provided by the above embodiment of the present invention is that at least one camera is arranged on the top, front and both sides of the sweeper, multiple laser radars are arranged around, at least one ultrasonic sensor is arranged on the front and rear bumpers, and at least one inertial sensor is arranged on the chassis; data information of each sensor is obtained and preprocessed, wherein the preprocessing includes data denoising and outlier removal; the data information of each sensor is characterized to obtain an approximate linear data set and a non-Gaussian data set; for the approximate linear data set, the prior probability distribution of the system state is estimated by the extended Kalman filter method, and the state estimation is corrected in combination with the latest observation data; for the non-Gaussian data set, random particles are generated by the particle filter method, the posterior probability distribution of the system is approximately estimated, and the weight of each particle is calculated according to the historical observation data, and the state estimation is obtained by weighted average; the fusion weights of the extended Kalman filter method and the particle filter method are dynamically adjusted by the adaptive fusion method to obtain the optimal fusion environment perception information; the fusion environment perception information is checked for consistency and redundancy, and the verified environment perception information is transmitted to the control system of the sweeper for navigation, obstacle avoidance and path planning. The present invention converts the research on fusion and verification of multi-source sensor information into the research on evaluation of sensor information, and improves the reliability of information fusion and verification by estimating data sets with different characteristics using extended Kalman filtering or particle filtering methods. Specifically, by setting cameras, laser radars, ultrasonic sensors and inertial sensors on multiple key parts such as the top, front, both sides and chassis of the sweeper, all-round perception of the surrounding environment is achieved, which not only ensures the comprehensiveness of information, but also improves the accuracy of information through the complementarity between different sensors. Data denoising uses the least squares fitting P-order polynomial method in the local data window to remove high-frequency noise and retain useful information. Outlier removal calculates the mean and standard deviation of data points, regards data points exceeding a certain threshold as outliers and removes them, further reducing the impact of noise and interference on information fusion. The extended Kalman filter accurately estimates the state of the system through prediction and update steps, and the particle filter approximates the posterior probability distribution of the system by generating random particles and calculating their weights. By estimating data sets with different characteristics using extended Kalman filtering or particle filtering methods, the reliability of information fusion and verification is further improved. By dynamically adjusting the fusion weights of the extended Kalman filter and the particle filter through an adaptive fusion method, the sweeper can intelligently select the fusion strategy that best suits the current environmental conditions, ensuring the accuracy and real-time nature of environmental perception information, and achieving optimal performance in both linear and nonlinear, Gaussian and non-Gaussian environments. In addition, consistency checks and redundancy checks are performed on the fused environmental perception information, further enhancing the reliability and robustness of the information and reducing the risk of misjudgment and misoperation. As a result, the sweeper is more efficient and safer in navigation, obstacle avoidance, and path planning.

FIG3 shows a schematic diagram of the structure of an embodiment of a computing device of the present invention. The specific embodiment of the present invention does not limit the specific implementation of the computing device.

As shown in FIG. 3 , the computing device may include: a processor 302 , a communications interface 304 , a memory 306 , and a communications bus 308 .

The processor 302, the communication interface 304, and the memory 306 communicate with each other via the communication bus 308. The communication interface 304 is used to communicate with other devices such as a client or other server network elements. The processor 302 is used to execute the program 310, which can specifically execute the relevant steps in the above-mentioned embodiment of the multi-source sensor information fusion and verification method for a road sweeper.

Specifically, the program 310 may include program codes, which include computer operation instructions.

The processor 302 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the computing device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

The memory 506 is used to store the program 310. The memory 306 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

The algorithm or display provided here is not inherently related to any specific computer, virtual system or other equipment. Various general systems can also be used together with the teaching based on this. According to the above description, it is obvious to construct the required structure of this type of system. In addition, the embodiment of the present invention is not directed to any specific programming language yet. It should be understood that various programming languages can be utilized to realize the content of the present invention described here, and the description of the above specific language is for disclosing the best mode of the present invention.

In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

Similarly, it should be understood that in order to streamline the present invention and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, the various features of the embodiments of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting the following intention: that the claimed invention requires more features than the features explicitly recited in each claim. More specifically, as reflected in the claims below, inventive aspects lie in less than all the features of the individual embodiments disclosed above. Therefore, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, with each claim itself serving as a separate embodiment of the present invention.

Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and further may be divided into multiple submodules or subunits or subcomponents. All features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device so disclosed may be combined in any combination, except that at least some of such features and/or processes or units are mutually exclusive. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature that provides the same, equivalent or similar purpose.

In addition, those skilled in the art will appreciate that, although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present invention and form different embodiments. For example, in the claims below, any one of the claimed embodiments may be used in any combination.

The various component embodiments of the present invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or a digital signal processor (DSP) may be used in practice to implement some or all of the functions of some or all of the components according to the embodiments of the present invention. The present invention may also be implemented as a device or apparatus program (e.g., a computer program and a computer program product) for executing part or all of the methods described herein. Such a program implementing the present invention may be stored on a computer-readable medium, or may be in the form of one or more signals. Such a signal may be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

It should be noted that the above embodiments illustrate the present invention rather than limit it, and that those skilled in the art may design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets shall not be construed as a limitation on the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention may be implemented by means of hardware comprising a number of different elements and by means of a suitably programmed computer. In a unit claim that lists a number of devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order. These words may be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be understood as limitations on the order of execution.

Claims (10)

1. A method for fusion and verification of multi-source sensor information of a road sweeper, characterized in that the method comprises: At least one camera is disposed on the top, front and both sides of the sweeper, a plurality of laser radars are disposed around, at least one ultrasonic sensor is disposed on the front and rear bumpers, and at least one inertial sensor is disposed on the chassis; Acquire data information from each sensor and perform preprocessing, wherein the preprocessing includes data denoising and outlier removal; The data information of each sensor is divided into characteristics to obtain an approximate linear data set and a non-Gaussian data set; for the approximate linear data set, the prior probability distribution of the system state is estimated by the extended Kalman filter method, and the state estimation is corrected in combination with the latest observation data; for the non-Gaussian data set, random particles are generated by the particle filter method to approximately estimate the posterior probability distribution of the system, and the weight of each particle is calculated according to the historical observation data, and the state estimation is obtained by weighted average; The fusion weights of the extended Kalman filter method and the particle filter method are dynamically adjusted through the adaptive fusion method to obtain the optimal fusion environment perception information; The fused environmental perception information is subjected to consistency check and redundancy check, and the checked environmental perception information is transmitted to the control system of the sweeper for use in navigation, obstacle avoidance and path planning. 2. The method according to claim 1, characterized in that the data denoising method further comprises: Preset the filter window size and the P-order polynomial order according to the raw data sequence of the multiple sensor data information; For each data point in the original data series ( ), select A local data window centered on the Nth dimension and containing N data points is used; a P-order polynomial is fitted on the local data window using the least squares method to obtain fitting polynomial coefficients; Use this fitting polynomial in Evaluate at and get the filtered data point ( ),Will The new filtered data sequence is combined in sequence; the filtered data sequence is used as the sensor data information after high-frequency noise is removed. 3. The method according to claim 1 or 2, wherein the method for removing outliers further comprises: According to the raw data sequences of multiple sensor data points, the mean value (μ) and standard deviation (σ) of all data points in the raw data sequences are calculated; For each data point in the original data series ( ), if | -μ| ≤ 3σ, then the data point is a normal value and is retained in the data series; if | - μ| > 3σ, then the data point is an outlier and is removed from the data sequence; The processed data sequence is used as the sensor data information after removing outliers. 4. The method according to claim 1, characterized in that the step of dividing the data information of each sensor into characteristics to obtain an approximately linear data set and a non-Gaussian data set further comprises: Calculate the P value of the data information of each sensor by the Anderson-Darling test method, and classify the data information with the P value greater than or equal to the preset significance level as a non-Gaussian data set; For data information with a P value less than the preset significance level, the R-square value is obtained by fitting the linear regression model, and the data information with an R-square value greater than or equal to the preset value is divided into an approximate linear data set. 5. The method according to claim 1, characterized in that estimating the prior probability distribution of the system state by the extended Kalman filter method and correcting the state estimation in combination with the latest observation data further comprises: Step S1, set the initial state estimate ( ) and the initial covariance matrix ( ); Step S2, according to the nonlinear state transfer function ( ) Predict the current state estimate ( ); Step S3, calculate the state transfer function ( )’s Jacobian matrix ( ), according to the Jacobian matrix ( ) and the process noise covariance matrix ( ) Prediction covariance( ); Step S4, according to the observation function ( )’s Jacobian matrix ( ) and the observation noise covariance matrix ( )Calculate the Kalman gain ( ); Step S5, according to the observed value ( ) and Kalman gain ( ) Update state estimate( ); Step S6, according to the Kalman gain ( ), Jacobian matrix( ) and covariance ( ) Update the covariance matrix ( ); Step S7, estimate the current state ( ) and the covariance matrix ( ) as the initial value for the next moment, and repeat steps S2 and S3. 6. The method according to claim 5, characterized in that, in step S2, the state estimation at the current moment ( ) is calculated as: in, is the state estimate at the previous moment, is the control input at the current moment; In step S3, the covariance ( ) is calculated as: in, is the covariance matrix of the state estimate at the previous moment obtained after updating the previous time step in the extended Kalman filter context; In step S4, the Kalman gain ( ) is calculated as: In step S5, the state estimation ( ) is calculated as: In step S6, the covariance matrix ( ) is calculated as: Where I is the identity matrix. 7. The method according to claim 1, characterized in that the generating random particles by a particle filtering method and approximating the posterior probability distribution of the system further comprises: Step S21, at time step (t=0), from the prior distribution ( ) and extract N particles ( ), assigning equal weight to each particle; Step S22, extract a sample for each particle from the importance distribution ( ); Step S23, calculate the weight of each particle according to the observation likelihood and state transition probability ( ); Step S24, estimating the posterior probability distribution of the system based on the particles and their weights. 8. The method according to claim 7, characterized in that, in step S23, the weight of each particle is: in, is the observation likelihood, indicating that given the state Observed The probability of is the state transition probability, which is from state Transfer to The state transition probability, N is the number of particles; In step S24, the calculation formula of the posterior probability distribution is: in, is the Dirac function, which represents the particle The position in state space, represents the state of the ith particle at time step t, is the normalized weight of the ith particle at time step t, Indicates that given historical data Under the condition of The probability distribution of is the data sequence observed by the sensor. 9. A multi-source sensor information fusion and verification device for a road sweeper, characterized by comprising: A setting module is used to set at least one camera on the top, front and both sides of the sweeper, set multiple laser radars around, set at least one ultrasonic sensor on the front and rear bumpers, and set at least one inertial sensor on the chassis; A preprocessing module, used to obtain data information from each sensor and perform preprocessing, wherein the preprocessing includes data denoising and outlier removal; A state estimation module is used to divide the data information of each sensor into characteristics to obtain an approximate linear data set and a non-Gaussian data set; for the approximate linear data set, the prior probability distribution of the system state is estimated by the extended Kalman filter method, and the state estimation is corrected in combination with the latest observation data; for the non-Gaussian data set, random particles are generated by the particle filter method to approximately estimate the posterior probability distribution of the system, and the weight of each particle is calculated according to the historical observation data, and the state estimation is obtained by weighted average; A fusion module is used to dynamically adjust the fusion weights of the extended Kalman filter method and the particle filter method through an adaptive fusion method to obtain the optimal fusion environment perception information; The verification module is used to perform consistency check and redundancy check on the fused environmental perception information, and transmit the verified environmental perception information to the control system of the sweeper for navigation, obstacle avoidance and path planning. 10. A computing device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the multi-source sensor information fusion and verification method for a sweeper according to any one of claims 1-8.