An automatic driving multi-source perception uncertainty evaluation method and device and medium

Abstract

The application relates to an automatic driving multi-source perception uncertainty evaluation method, which comprises the following steps: obtaining a target detection result of target detection processing on sensing data of a multi-source sensor; establishing an occupancy grid map based on sensing data of one kind of sensor, and mapping the target detection result corresponding to the multi-source sensor on the occupancy grid map to obtain corresponding perception results of each sensor on the occupancy grid map; performing matching processing on the target detection result corresponding to each sensor according to a target matching algorithm to obtain a target matching result; and performing statistical analysis on the corresponding perception result of each sensor on the occupancy grid map and the target matching result to obtain an evaluation result of perception uncertainty. The application can evaluate the uncertainty of the perception result in real time, efficiently and accurately, and guarantee the driving safety of automatic driving.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving environmental perception technology, and in particular to a method, apparatus, and computer-readable storage medium for assessing uncertainties in multi-source perception for autonomous driving. Background Technology

[0002] Autonomous driving perception enables the acquisition of environmental information in real time, serving as a crucial information source for the safe and efficient operation of autonomous driving systems. The accuracy and completeness of this data directly impact the functional implementation of downstream modules, ultimately affecting the safety and efficiency of real-time autonomous driving operations.

[0003] With the development of computer hardware, big data technology, and artificial intelligence, deep neural networks now play a crucial role in autonomous driving perception. However, the uninterpretable nature of the black-box mechanism of deep neural networks inevitably leads to uncertainties in the results of autonomous driving perception. The output of deep neural network algorithms can contain missed detections, false detections, positional errors, dimensional errors, and orientation errors. If these errors are not detected in time, they may cause accidents involving autonomous driving systems, endangering personal safety. Therefore, real-time evaluation of the performance of autonomous driving perception results and quantification of perception uncertainty information are of great significance for improving the safety of autonomous driving.

[0004] Currently, research on perception mainly focuses on improving perception performance, while systematic research on the quantitative assessment of uncertainty in autonomous driving perception is still immature. Relevant literature primarily focuses on the uncertainty of deep neural network algorithms, including the sources of uncertainty, classification of uncertainty, assessment methods, assessment metrics, and responses to perception uncertainty. However, most of these methods are based on offline evaluation using labeled dataset values, while the few online evaluation methods, represented by deep ensemble methods, can meet the requirements of real-time performance and accuracy. However, these methods struggle to cover data distributions outside the training network when assessing uncertainty. In conclusion, how to assess the uncertainty of perception results in real-time, efficiently, and accurately to ensure the driving safety of autonomous driving remains an unresolved issue. Summary of the Invention

[0005] To address the aforementioned problems, the present invention aims to provide a method, apparatus, and computer-readable storage medium for assessing uncertainties in multi-source perception for autonomous driving, which can assess the uncertainty of perception results in real time, efficiently, and accurately, thereby ensuring driving safety in autonomous driving.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] Firstly, this application provides a method for assessing the uncertainty of multi-source perception in autonomous driving, the method comprising:

[0008] Obtain target detection results by processing sensor data from multiple sources;

[0009] Based on the sensing data of one of the sensors, an occupied grid map is established, and the target detection results corresponding to the multi-source sensors are mapped onto the occupied grid map to obtain the perception results of each sensor on the occupied grid map.

[0010] Based on the target matching algorithm, the target detection results corresponding to each sensor are matched to obtain the target matching result;

[0011] Based on the perception results of each sensor on the occupied grid map and the target matching results, statistical analysis is performed to obtain the evaluation results of perception uncertainty.

[0012] In one implementation of this application, the multi-source sensor includes a lidar and a camera;

[0013] The step of establishing an occupied grid map based on sensing data from one of the sensors includes: establishing an occupied grid map based on the original point cloud acquired by LiDAR.

[0014] In one implementation of this application, the step of establishing an occupancy grid map based on the original point cloud acquired by lidar includes:

[0015] The ground point cloud in the original point cloud is filtered;

[0016] The filtered ground point cloud is separated into discrete two-dimensional toruses formed by grids in spherical coordinates, with each discrete two-dimensional torus containing multiple grids.

[0017] In one implementation of this application, the target detection result corresponding to the multi-source sensor includes the position of the detected target and the probability of semantic classification;

[0018] The step of mapping the target detection results corresponding to the multi-source sensors onto the occupied grid map to obtain the perception results of each sensor on the occupied grid map includes:

[0019] The target detection results corresponding to the multi-source sensors are mapped to obtain the occupancy probability and semantic classification probability of each grid corresponding to each sensor on the occupancy grid map.

[0020] In one implementation of this application, the mapping process includes:

[0021] The grid state O at time k k Defined as a binary opposition: occupied or free, i.e., O k ∈{O,F};

[0022] Based on the measurement z at time k+1 k+1 To estimate the predicted occupancy probability for each grid cell

[0023] The posterior probability of each grid cell is recursively estimated using Bayesian rules, calculated as follows:

[0024]

[0025] in, This represents the probability of occupying grid O in a raster map. This represents the probability that grid O is in a free state.

[0026] In one implementation of this application, the step of matching the target detection results corresponding to each sensor according to a target matching algorithm to obtain the target matching result includes:

[0027] The matching process for target detection results is implemented based on the K-nearest neighbor algorithm and the triangular total error matching algorithm.

[0028] In one implementation of this application, the evaluation result of the perceived uncertainty includes:

[0029] The assessment results of the existence uncertainty of the target, the semantic uncertainty assessment results, and the spatial uncertainty assessment results.

[0030] In one implementation of this application, the evaluation result with uncertainty is expressed as follows:

[0031]

[0032] In the formula, This represents the probability that the grid in the i-th row and j-th column of the region occupied by the detected target is not occupied by the original point cloud of the LiDAR. This represents the probability that the grid in the i-th row and j-th column of the region occupied by the detected target is not occupied by the perceived target;

[0033] The semantic uncertainty assessment result uses the prediction entropy E pe To represent this, the calculation formula can be expressed as:

[0034] E pe =-(P c ·logP c +(1-P c )·log(1-P c ))

[0035] Wherein, the occupancy probability of a grid is P0, and the semantic probability of each grid is P. c ;

[0036] The spatial uncertainty assessment results are represented by the uncertainty of the target's position and size:

[0037] TV L =var x +var y +var z

[0038] TV D =var l +var w +var h

[0039] Shizhong TV L This represents the uncertainty in the position of the perceived target, including the uncertainty in the target's coordinates in the horizontal, vertical (y), and z-direction directions. (TV) D This indicates the uncertainty in the size of the perceived target, including the uncertainty in the length, width, and height of the perceived target.

[0040] Secondly, this application provides an autonomous driving multi-source perception uncertainty assessment device, characterized in that the device comprises:

[0041] The target acquisition unit is used to acquire the target detection results by processing the sensing data from multiple source sensors.

[0042] The mapping unit is used to establish an occupied grid map based on the sensing data of one of the sensors, and to map the target detection results corresponding to the multi-source sensors onto the occupied grid map to obtain the perception results of each sensor on the occupied grid map.

[0043] The target matching unit is used to perform matching processing on the target detection results corresponding to each sensor according to the target matching algorithm to obtain the target matching result;

[0044] The evaluation unit is used to perform statistical analysis based on the perception results of each sensor on the occupied grid map and the target matching results to obtain the evaluation result of perception uncertainty.

[0045] Thirdly, this application provides a computer-readable storage medium storing a computer program, wherein the computer program, when running, controls the device where the computer-readable storage medium is located to execute the autonomous driving multi-source perception uncertainty assessment method described in the first aspect.

[0046] The present invention has the following advantages due to the adoption of the above technical solutions: The present invention obtains the target detection results of the sensing data of multiple source sensors through target detection processing, and then establishes an occupied grid map based on the sensing data of one of the sensors. The target detection results corresponding to the multiple source sensors are mapped onto the occupied grid map to obtain the perception results corresponding to each sensor on the occupied grid map. Furthermore, according to the target matching algorithm, the target detection results corresponding to each sensor are matched to obtain the target matching results. Based on the perception results and target matching results corresponding to each sensor on the occupied grid map, statistical analysis is performed to obtain the evaluation results of perception uncertainty, thereby enabling real-time, efficient and accurate evaluation of the uncertainty of perception results. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating an uncertainty assessment method for multi-source perception in autonomous driving provided by an embodiment of the present invention.

[0048] Figure 2 This is a flowchart of the evaluation process for perceived uncertainty in a detailed embodiment of the present invention;

[0049] Figure 3 This is a flowchart of the target matching algorithm based on KNN and the total error of the triangle in an embodiment of the present invention;

[0050] Figure 4 This is a schematic diagram of triangle matching used in an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0052] To address the problem of the urgent need for real-time, efficient, and accurate assessment of the uncertainty of perception results in existing technologies, this invention provides a method, apparatus, and computer-readable storage medium for assessing the uncertainty of multi-source perception in autonomous driving. The method includes: acquiring target detection results from target detection processing of sensor data from multiple sources; establishing an occupied grid map based on sensor data from one of the sensors, and mapping the target detection results corresponding to the multiple sources onto the occupied grid map to obtain the perception results corresponding to each sensor on the occupied grid map; performing matching processing on the target detection results corresponding to each sensor according to a target matching algorithm to obtain target matching results; and performing statistical analysis based on the perception results corresponding to each sensor on the occupied grid map and the target matching results to obtain an assessment result of the perception uncertainty. This solution can assess the uncertainty of perception results in real-time, efficiently, and accurately, ensuring driving safety in autonomous driving.

[0053] Please refer to the accompanying drawings of the embodiments of the present invention for further details on the methods, apparatus and media provided by the present invention in the more detailed embodiments of the present invention.

[0054] like Figure 1 In one aspect of the embodiments of this application, a method for evaluating the uncertainty of multi-source perception in autonomous driving is provided, specifically including:

[0055] S01, Obtain the target detection result by processing the sensing data from the multi-source sensors;

[0056] S02, Based on the sensing data of one of the sensors, an occupied grid map is established, and the target detection results corresponding to the multi-source sensors are mapped onto the occupied grid map to obtain the perception results of each sensor on the occupied grid map.

[0057] S03, According to the target matching algorithm, the target detection results corresponding to each sensor are matched to obtain the target matching result;

[0058] S04. Based on the perception results of each sensor on the occupied grid map and the target matching results, perform statistical analysis to obtain the evaluation result of perception uncertainty.

[0059] The following is combined Figures 2 to 4 In a more detailed embodiment of this application, the method provided in this application is described.

[0060] The method includes:

[0061] S01, Obtain the target detection result by processing the sensing data from the multi-source sensors;

[0062] Specifically, such as Figure 2 As shown, the multi-source sensors in this embodiment include LiDAR and a camera. This application may also include more types or numbers of sensors.

[0063] Figure 2 In this process, a camera-based perception algorithm is used to perform target detection on the camera's sensing data, i.e., the image, to obtain the target detection result corresponding to the camera.

[0064] A LiDAR-based perception algorithm is used to perform target detection on the LiDAR sensing data, i.e., point cloud, to obtain the target detection results corresponding to the LiDAR.

[0065] The two perception algorithms mentioned above can be models such as neural networks. The target detection result can include the location of the detected target and the probability of semantic classification. The semantic classification result can be a pre-defined detection category such as a person, vehicle, or obstacle.

[0066] S02, Based on the sensing data of one of the sensors, an occupied grid map is established, and the target detection results corresponding to the multi-source sensors are mapped onto the occupied grid map to obtain the perception results of each sensor on the occupied grid map.

[0067] Specifically, in this embodiment, an Occupied Grid Map (OGM) is established based on the original point cloud acquired by the lidar.

[0068] This embodiment utilizes the Random Sample Consensus Algorithm (RANSAC) to filter the ground point cloud. The filtered ground lidar point cloud is separated into discrete two-dimensional tori formed by a grid in spherical coordinates.

[0069] In this embodiment of the application, the mapping process is to map the target detection results corresponding to the multi-source sensors onto the OGM to obtain the occupancy probability and semantic classification probability of each sensor on the OGM.

[0070] The calculation of the occupancy probability follows the following process:

[0071] The grid state O at time k k Defined as a binary opposition: occupied (O) or free (F), i.e., O k ∈{O,F}; Over time, sensor measurements are used to update the state O of the grid cells. k This model, which uses distance measurement and vehicle configuration to obtain the grid occupancy probability, is called the inverse sensor model.

[0072] The inverse sensor model is based on the measurement z at time k+1. k+1 To estimate the predicted occupancy probability for each grid cell Using this measurement as an estimate, the posterior occupancy probability P at time k is... o,k (O k Update the occupancy probability P at time k+1. o,k+1 (O k+1 );

[0073] The posterior probability of each grid cell is recursively estimated using Bayesian rules, calculated as follows:

[0074]

[0075] Wherein, P(O), This represents the probability of occupying grid O in a raster map. This represents the probability that grid O is in a free state.

[0076] Mapping the multi-source perception results onto the occupied grid map yields the perception results from the Bird's Eye View (BEV) perspective.

[0077] S03, According to the target matching algorithm, the target detection results corresponding to each sensor are matched to obtain the target matching result;

[0078] Because of the spatial uncertainty of the perceived target, the results of multi-source sensing occupying a grid map are often mismatched. Therefore, a target matching algorithm is needed.

[0079] This embodiment uses a matching algorithm based on the K-Nearest Neighbor (KNN) algorithm and the total triangle error to achieve matching of different detection targets. The specific process is as follows: Figure 3 As shown, it can be summarized as follows:

[0080] (1) Set the target set S with matching A and target set S B (e.g. S) A Corresponding to the target detection results of the camera, S B (corresponding to the target detection results of the lidar), the number of targets is N. A and N B ;

[0081] (2) Calculate the minimum number of objects identified in each frame from both sources, m = min{N A N B}

[0082] (3) If m = 1, then use the K-nearest neighbor algorithm to calculate the target set S. A All object points to set S BThe distances between all object points. The formula for calculating the distance between target point A and target point B can be expressed as follows:

[0083]

[0084] Where, d AB This represents the three-dimensional spatial distance between target points, in meters (m); a distance threshold τ is set. d If d AB <τ d If the two target points match, it means they are compatible. Otherwise, they are not compatible, indicating a possible missed detection or false detection.

[0085] (4) If m = 2, in order for the target points in the perception results to form a triangle, a virtual target point needs to be added at a certain distance. Generally, the range of the perception algorithm is set within 50m, so virtual target detection points are usually set at a relatively far distance. For example, in this invention, the virtual points selected from the two perception sources are (500m, 500m) and (499m, 499m). It should be noted that the virtual points from different perception sources cannot be completely identical, but the distance cannot be too far. After adding the virtual target point, the triangle matching algorithm can be used.

[0086] (5) If m is greater than or equal to 3, then the triangle matching method is used.

[0087] (6) The triangle matching method can be described as a target from the BEV perspective, therefore it consists of two-dimensional coordinates: horizontal (x) and vertical (z), including:

[0088] The first step is to analyze S. A S B Each point is numbered;

[0089] The second step is to solve S separately. A S B The difference between the maximum and minimum values ​​of x and z can be calculated using the following formula:

[0090]

[0091]

[0092] The third step is to treat S A S B Sort all points in the triangle: find the maximum distance difference. If it is an index element of the x-coordinate, sort by x; otherwise, sort by z. This ensures that the points correspond one-to-one in this order when matching triangles.

[0093] Fourth step, for S A S BRandomly select three points from all the points in the array to form a triangle, and return the index of the triangle's location.

[0094] Fifth step, for S A S B The triangles formed by all points in the array are normalized by dividing the side length of each triangle by the shortest side length; this ensures the setting of the threshold for successful matching.

[0095] Step 6, for S A Each triangle in the diagram is associated with S. B Find the sum of the side errors and the sum of the point errors for all triangles in the given triangle. Take any two points in the triangle as an example... Figure 4 As shown, points A1 and A2 are corresponding points:

[0096]

[0097] In the formula, E P This represents the error at a point. The error E for an edge... E It can be represented as:

[0098]

[0099] The total error between the two triangles can be expressed as:

[0100]

[0101] Step 7: Find the minimum value E among all triangular errors. Tri_max If E Tri_max Less than the triangle error threshold τ T If the two triangles are matched, then the points of the triangles after the sorting in the third step above will be the target of the match.

[0102] Step 8: Analyze the remaining points, and then repeat steps (3), (4), and (5) until all points are matched.

[0103] (7) Output the matching points, which are the points detected normally; if S A Consider it as the set of truth values, S B For the set to be detected, then in S A In S B In S, it is considered a missed detection; B In S A The ones in the middle are considered false inspections.

[0104] S04. Based on the perception results of each sensor on the occupied grid map and the target matching results, perform statistical analysis to obtain the evaluation result of perception uncertainty.

[0105] The assessment results of perceived uncertainty include: assessment results of the existence uncertainty of the target, assessment results of semantic uncertainty, and assessment results of spatial uncertainty.

[0106] After multi-source sensing target matching, a target from one sensing source may be missed or falsely detected. To further determine which type of existence uncertainty it belongs to, verification based on the original point cloud is required. Therefore, the sensing result is matched with the original point cloud on the occupied grid map. If the entire grid at the same location is occupied, it is considered a missed detection. Otherwise, the sensing result contains a false detection. Assuming the area occupied by the detected object on the occupied grid map is A, the sensing judgment result is expressed as:

[0107]

[0108] In the formula, This represents the probability that the grid in the i-th row and j-th column of the region occupied by the detected object is not occupied by the original point cloud of the LiDAR. This represents the probability that the grid in the i-th row and j-th column of the region occupied by the detected object is not occupied by the perceived target. These occupancy probabilities are calculated using the grid occupancy probability calculation formula mentioned above, and are the posterior occupancy probabilities at the current time derived from Bayesian filtering.

[0109] Uncertainty assessment mainly includes semantic uncertainty and spatial uncertainty of the perceived target. After the multi-source sensing results are mapped to the occupied grid map, if a grid is occupied, the occupancy probability of the corresponding grid is P0, and the semantic probability of each grid is P. c The semantic uncertainty of each grid is expressed using the prediction entropy E. pe To represent this, the calculation formula can be expressed as:

[0110] E pe =-(P c ·logP c +(1-P c )·log(1-P c ))

[0111] Furthermore, after matching the results of multi-source sensing, the uncertainty of the matched target's position, size, orientation, and other information can be calculated. The spatial information of the sensed target includes the detected object's lateral position (x), longitudinal position (z), vertical position (y), length (l), width (w), and height (h), as well as its orientation (r). The corresponding uncertainty evaluation indicators can be represented by variance and total variance (TV).

[0112]

[0113] Similarly, the variances of other metrics are calculated using the same equations. Each object has a position and a size, and the TV of position and size can be calculated using the following formula.

[0114] TV L =var x +var y +var z

[0115] TV D =var l +var w +var h

[0116] Shizhong TV L This indicates the uncertainty in the position of the perceived target, including the uncertainty in the target's coordinates in the horizontal, vertical, and longitudinal directions. (TV) D This represents the uncertainty in the size of the perceived target, including the uncertainty in its length, width, and height. The above completes the calculation of the spatial uncertainty of the perceived target.

[0117] Another aspect of this application embodiment also provides a corresponding apparatus, including:

[0118] The target acquisition unit is used to acquire the target detection results by processing the sensing data from multiple source sensors.

[0119] The mapping unit is used to establish an occupied grid map based on the sensing data of one of the sensors, and to map the target detection results corresponding to the multi-source sensors onto the occupied grid map to obtain the perception results of each sensor on the occupied grid map.

[0120] The target matching unit is used to perform matching processing on the target detection results corresponding to each sensor according to the target matching algorithm to obtain the target matching result;

[0121] The evaluation unit is used to perform statistical analysis based on the perception results of each sensor on the occupied grid map and the target matching results to obtain the evaluation result of perception uncertainty.

[0122] In another aspect of the embodiments of this application, a computer storage medium is also provided.

[0123] This application also provides a computer-readable storage medium, which includes a stored program, wherein the program, when running, controls the device where the storage medium is located to execute the above-described method. The specific implementation process will not be repeated here.

[0124] This application also provides a computer device. The computer device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the methods described above in the embodiment; to avoid repetition, these will not be elaborated further here. Alternatively, when the computer program is executed by the processor, it implements the functions of each model / unit in the device of the embodiment; to avoid repetition, these will not be elaborated further here.

[0125] Computer equipment can be desktop computers, laptops, handheld computers, servers, and cloud servers, etc. Computer equipment may include, but is not limited to, processors and memory. Those skilled in the art will understand that it may include more or fewer components than illustrated, or combine certain components, or different components; for example, computer equipment may also include input / output devices, network access devices, buses, etc.

[0126] The processor referred to can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0127] Memory can be an internal storage unit of a computer device, such as a hard drive or RAM. Memory can also be an external storage device of a computer device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, memory can include both internal and external storage units. Memory is used to store computer programs and other programs and data required by the computer device. Memory can also be used to temporarily store data that has been output or will be output.

[0128] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0129] In the embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0130] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0131] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for assessing uncertainty in multi-source perception for autonomous driving, characterized in that, The method includes: Obtain target detection results by processing sensor data from multiple sources; Based on the sensing data of one of the sensors, an occupied grid map is established, and the target detection results corresponding to the multi-source sensors are mapped onto the occupied grid map to obtain the perception results of each sensor on the occupied grid map. Based on the target matching algorithm, the target detection results corresponding to each sensor are matched to obtain the target matching result; Based on the perception results of each sensor on the occupied grid map and the target matching results, statistical analysis is performed to obtain the evaluation results of perception uncertainty; The multi-source sensors include lidar and cameras; the step of establishing an occupied grid map based on the sensing data of one of the sensors includes: establishing an occupied grid map based on the original point cloud acquired by lidar. The process of establishing an occupied grid map based on the original point cloud acquired by lidar includes: filtering the ground point cloud in the original point cloud; and separating the filtered ground point cloud into discrete two-dimensional toruses formed by grids in spherical coordinates, with each discrete two-dimensional torus containing multiple grids. The target detection results corresponding to the multi-source sensors include the location of the detected target and the probability of semantic classification; the step of mapping the target detection results corresponding to the multi-source sensors onto the occupied grid map to obtain the perception results of each sensor on the occupied grid map includes: performing mapping processing on the target detection results corresponding to the multi-source sensors to obtain the occupancy probability and semantic classification probability of each grid corresponding to each sensor on the occupied grid map; The assessment results of perceived uncertainty include: assessment results of the existence uncertainty of the target, assessment results of semantic uncertainty, and assessment results of spatial uncertainty; After multi-source sensing target matching, a target from a particular sensing source may be missed or falsely detected. To further determine which type of existence uncertainty it belongs to, verification based on the original point cloud is required. Therefore, the sensing result is matched with the original point cloud occupying the grid map. If all grids at the same location are occupied, it is considered a missed detection; otherwise, the sensing result contains a false detection. The assessment result of the existence uncertainty is expressed as follows: = In the formula, Indicates the number of regions occupied by the detected target. i Line number j The probability that the grid of the column is not occupied by the original point cloud of the LiDAR. Indicates the number of regions occupied by the detected target. i Line number j The probability that the grid of a column is not occupied by a perceived target; The semantic uncertainty assessment result uses prediction entropy. To represent this, the calculation formula can be expressed as: Wherein, the occupancy probability of the grid is The semantic probability of each grid is ; The spatial uncertainty assessment results are represented by the uncertainty of the target's position and size: In the formula This represents the uncertainty in the position of the perceived target, including the uncertainty in the target's coordinates in the horizontal, vertical (y), and z-direction directions. This indicates the uncertainty in the size of the perceived target, including the uncertainty in the length, width, and height of the perceived target.

2. The method for assessing uncertainty in multi-source perception for autonomous driving according to claim 1, characterized in that, The mapping process includes: time k Grid status at the location Defined as a binary opposition: occupied or free, that is ; Based on time k +1 measurement value To estimate the predicted occupancy probability for each grid cell ; The posterior probability of each grid cell is recursively estimated using Bayesian rules, calculated as follows: in, , Indicates occupying a grid in a raster map The probability of occupying a state. Represents a grid The probability of being in a free state.

3. The method for assessing uncertainty in multi-source perception for autonomous driving according to claim 2, characterized in that, The step of matching the target detection results corresponding to each sensor according to the target matching algorithm to obtain the target matching result includes: The matching process for target detection results is implemented based on the K-nearest neighbor algorithm and the triangular total error matching algorithm.

4. An autonomous driving multi-source perception uncertainty assessment device, used to implement the autonomous driving multi-source perception uncertainty assessment method as described in any one of claims 1-3, characterized in that, The device includes: The target acquisition unit is used to acquire the target detection results by processing the sensing data from multiple source sensors. The mapping unit is used to establish an occupied grid map based on the sensing data of one of the sensors, and to map the target detection results corresponding to the multi-source sensors onto the occupied grid map to obtain the perception results of each sensor on the occupied grid map. The target matching unit is used to perform matching processing on the target detection results corresponding to each sensor according to the target matching algorithm to obtain the target matching result; The evaluation unit is used to perform statistical analysis based on the perception results of each sensor on the occupied grid map and the target matching results to obtain the evaluation result of perception uncertainty.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed, controls the device containing the computer-readable storage medium to perform the autonomous driving multi-source perception uncertainty assessment method according to any one of claims 1 to 3.

Images