Vehicle track updating method, device, equipment, medium and program product

Abstract

The invention provides a vehicle track updating method, device and equipment, a medium and a program product. The method comprises the following steps: acquiring a visual obstacle and a plurality of radar obstacles corresponding to a vehicle, and screening the plurality of radar obstacles to obtain a plurality of first candidate obstacles; determining an association probability between each first candidate obstacle and the visual obstacle according to the attribute information of the plurality of first candidate obstacles and the attribute information of the visual obstacle, and determining a plurality of second candidate obstacles based on the association probabilities; determining a first distance parameter of each second candidate obstacle, and determining a target obstacle from the plurality of second candidate obstacles based on the first distance parameters; and updating the track of the vehicle based on the target obstacle. According to the scheme, the track of the vehicle is updated by associating the visual obstacle with the radar obstacle and calculating the first distance parameter to screen out the target obstacle; therefore, the track stability and correctness of the vehicle are improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a vehicle track updating method, device, equipment, medium and program product. Background Art

[0002] With the development of autonomous driving technology, the availability of assisted driving features has become a major consideration for vehicle purchasers. The key to autonomous driving technology lies in its ability to quickly and accurately perceive the environment and consistently provide effective information about objects around the vehicle. However, at long distances, the ability to accurately determine position and velocity attributes can be limited due to environmental factors and the difficulty in clearly capturing details of the target object.

[0003] Currently, the method for updating a vehicle's track is to collect and process image or video information, determine the attributes of position and speed, and then identify obstacles to update the vehicle's track; however, this method cannot ensure the stability and accuracy of the vehicle's track. Summary of the Invention

[0004] The embodiments of the present application provide a vehicle track updating method, apparatus, device, medium and program product for improving the stability and accuracy of the vehicle track.

[0005] In a first aspect, an embodiment of the present application provides a vehicle track updating method, comprising: obtaining a visual obstacle and multiple radar obstacles corresponding to the vehicle, screening the multiple radar obstacles to obtain multiple first candidate obstacles; determining an association probability between each first candidate obstacle and the visual obstacle based on attribute information of the multiple first candidate obstacles and attribute information of the visual obstacle, and determining multiple second candidate obstacles based on the association probability; determining a first distance parameter for each second candidate obstacle, and determining a target obstacle from the multiple second candidate obstacles based on the first distance parameter; and updating the vehicle track based on the target obstacle.

[0006] In some examples, updating a radar obstacle for a vehicle's track based on a target obstacle includes: determining whether the target obstacle is a currently associated target obstacle; if the target obstacle is not a currently associated target obstacle, associating the target obstacle with the vehicle's track to obtain a new track, and updating the association between the new track and the visual obstacle; filtering and smoothing attributes of the new track; and updating the vehicle's track to the new track; if the target obstacle is a currently associated target obstacle, associating the target obstacle with the vehicle's track to obtain a new track, and updating the association between the new track and the visual obstacle; and updating the vehicle's track to the new track.

[0007] In some examples, a plurality of radar obstacles are screened to obtain a plurality of first candidate obstacles, including: determining, for any one of the plurality of radar obstacles, a longitudinal distance error and a lateral distance error between the radar obstacle and a visual obstacle, and determining a distance error value corresponding to the radar obstacle based on the longitudinal distance error and the lateral distance error; determining whether the distance error value is less than a preset error value; if the distance error value is less than the preset error value, selecting the radar obstacle as a first candidate obstacle; and if the distance error value is not less than the preset error value, eliminating the radar obstacle.

[0008] In some examples, based on the longitudinal distance error and the lateral distance error, determining the distance error value corresponding to the radar obstacle includes: dividing the longitudinal distance error by a longitudinal preset parameter and squaring the result to obtain the longitudinal distance error value; dividing the lateral distance error by a lateral preset parameter and squaring the result to obtain the lateral distance error value; and summing the longitudinal distance error value and the lateral distance error value to obtain the distance error value corresponding to the radar obstacle.

[0009] In some examples, determining an association probability between each first candidate obstacle and the visual obstacle based on the attribute information of the first candidate obstacle and the attribute information of the visual obstacle, and determining multiple second candidate obstacles based on the association probabilities includes: for any first candidate obstacle among the multiple first candidate obstacles, determining whether the attribute information of the first candidate obstacle and the attribute information of the visual obstacle have a preset number of consistent attribute items; if the attribute consistency reaches the preset number of items, the association probability of the first candidate obstacle is non-zero, and the first candidate obstacle with a non-zero association probability is used as the second candidate obstacle; if the attribute consistency does not reach the preset number of items, the association probability of the first candidate obstacle is zero.

[0010] In some examples, determining the first distance parameter of each second candidate obstacle includes: determining, for any one of the plurality of second candidate obstacles and the visual obstacle, a state matrix of the visual obstacle and a state matrix of the second candidate obstacle, and determining the second distance parameter of the second candidate obstacle based on the state matrix of the visual obstacle and the state matrix of the second candidate obstacle; wherein the state matrix represents position data and velocity data of the second candidate obstacle and the visual obstacle; adjusting preset weights for different directions based on historical experience to obtain a first weight; wherein the different directions include longitudinal position, lateral position, longitudinal velocity, and lateral velocity; calculating a third distance parameter based on the second distance parameter and the first weight; obtaining a number of times the second candidate obstacle is associated, and calculating the first distance parameter based on the third distance parameter and the number of times the second candidate obstacle is associated using a nonlinear activation function; and selecting a minimum distance parameter from the first distance parameters.

[0011] In some examples, determining a second distance parameter of the second candidate obstacle based on the state quantity matrix of the visual obstacle and the state quantity matrix of the second candidate obstacle includes: subtracting the state quantity matrix of the visual obstacle from the state quantity matrix of the second candidate obstacle, transposing the resulting matrix to obtain a first transposed matrix of the visual obstacle and the second candidate obstacle; obtaining an inverse matrix of a first covariance matrix of the visual obstacle and the second candidate obstacle based on the first transposed matrix; and multiplying the first transposed matrix by the inverse matrix of the first covariance matrix to obtain the second distance parameter.

[0012] In some examples, a third distance parameter is obtained by performing a calculation based on the second distance parameter and the first weight, including: subtracting the state matrix of the visual obstacle from the state matrix of the second candidate obstacle, multiplying the obtained matrix by the first weight, and then transposing the matrix to obtain a second transposed matrix of the visual obstacle and the second candidate obstacle; obtaining an inverse matrix of a second covariance matrix between the visual obstacle and the second candidate obstacle based on the second transposed matrix, and multiplying the second transposed matrix by the inverse matrix of the second covariance matrix to obtain the third distance parameter.

[0013] In some examples, the first distance parameter is obtained by calculating based on the third distance parameter and the number of times the second candidate obstacle is associated using a nonlinear activation function, including: subtracting the number of times the second candidate obstacle is associated and dividing it by a preset parameter, and calculating the result using a nonlinear activation function to obtain an optimized value; subtracting the preset value from the optimized value, and multiplying the result by the third distance parameter to obtain the first distance parameter.

[0014] In some examples, the method further includes: determining whether a third distance parameter corresponding to the second candidate obstacle corresponding to the minimum distance parameter is a minimum value; if the corresponding third distance parameter is not a minimum value; updating the number of associations of the second candidate obstacle corresponding to the minimum distance parameter according to an exponential function, recalculating the first distance parameter and performing screening, and selecting the second candidate obstacle corresponding to the minimum distance parameter obtained by screening as the target obstacle; and if the corresponding third distance parameter is a minimum value, selecting the second candidate obstacle corresponding to the minimum distance parameter as the target obstacle.

[0015] In some examples, updating the number of times the second candidate obstacle is associated with the minimum distance parameter according to an exponential function includes: adding one to the number of association errors for the second candidate obstacle corresponding to the minimum distance parameter, taking the exponential value, multiplying the result by the number of times the second candidate obstacle is associated with the second candidate obstacle, and rounding down to obtain a new number of times the second candidate obstacle is associated.

[0016] In some examples, after the radar association probability of the first candidate obstacle is zero, the method further includes: using the first candidate obstacle with the zero association probability as a third candidate obstacle, and returning the third candidate obstacle to execute the above steps starting from screening multiple radar obstacles.

[0017] In a second aspect, an embodiment of the present application provides a vehicle track updating device, comprising: an acquisition module for acquiring a visual obstacle and multiple radar obstacles corresponding to the vehicle, screening and processing the multiple radar obstacles to obtain multiple first candidate obstacles; a processing module for determining an association probability between each first candidate obstacle and the visual obstacle based on attribute information of the multiple first candidate obstacles and attribute information of the visual obstacle, and determining multiple second candidate obstacles based on the association probability; the processing module is further configured to determine a first distance parameter for each second candidate obstacle, and determine a target obstacle from the multiple second candidate obstacles based on the first distance parameter; and the processing module is further configured to update the vehicle track based on the target obstacle.

[0018] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a memory, a processor;

[0019] The memory stores computer-executable instructions;

[0020] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the above first aspect and / or various possible implementations of the first aspect.

[0021] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the first aspect above and / or various possible implementation methods of the first aspect.

[0022] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the above first aspect and / or various possible implementation methods of the first aspect.

[0023] Embodiments of the present application provide a vehicle track updating method, apparatus, device, medium, and program product. The method first obtains visual obstacles and multiple radar obstacles corresponding to the vehicle, then screens and processes the multiple radar obstacles to obtain multiple first obstacle candidates. The method then determines the association probability between each first obstacle candidate and the visual obstacle based on the attribute information of the multiple first obstacle candidates and the attribute information of the visual obstacle. Based on the association probability, multiple second obstacle candidates are screened and determined. For each of the obtained second obstacle candidates, a first distance parameter is calculated, and based on the first distance parameter, a target obstacle is determined from the multiple second candidate obstacles. The vehicle track is then updated based on the target obstacle. This solution updates the vehicle track by associating visual obstacles with radar obstacles and calculating the first distance parameter to screen out the target obstacle, thereby improving the stability and accuracy of the vehicle track. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0025] Figure 1 A schematic diagram of the process of updating the vehicle track provided in this application Figure 1 ;

[0026] Figure 2 is a schematic diagram of a radar obstacle;

[0027] Figure 3 A schematic diagram of the process of updating the vehicle track provided in this application Figure 2 ;

[0028] Figure 4 Select a schematic diagram for a target obstacle;

[0029] Figure 5 Schematic diagram of the process of a vehicle track update method provided by this application Figure 3 ;

[0030] Figure 6 A schematic structural diagram of a vehicle track updating device provided in this application;

[0031] Figure 7 This is a schematic diagram of the structure of an electronic device provided in this application.

[0032] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0033] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0034] With the development of autonomous driving technology, the availability of assisted driving features has become a major consideration for vehicle purchasers. The key to autonomous driving technology lies in its ability to quickly and accurately perceive the environment and consistently provide effective information about objects around the vehicle. However, at long distances, the ability to accurately determine position and velocity attributes can be limited due to environmental factors and the difficulty in clearly capturing details of the target object.

[0035] Currently, the method for updating a vehicle's track is to collect and process image or video information, determine the attributes of position and speed, and then identify obstacles to update the vehicle's track; however, this method cannot ensure the stability and accuracy of the vehicle's track.

[0036] This application provides a vehicle track updating method. This method obtains visual obstacles and multiple radar obstacles corresponding to the vehicle, screens the multiple radar obstacles, and obtains multiple first obstacle candidates. The method then determines the association probability between each first obstacle candidate and the visual obstacle based on the attribute information of the multiple first obstacle candidates and the attribute information of the visual obstacle. Multiple second obstacle candidates are then identified based on the association probability. A first distance parameter is determined for each second obstacle candidate, and based on the first distance parameter, a target obstacle is identified from the multiple second obstacle candidates. The method then updates the vehicle track based on the target obstacle. This method solves the technical problem of difficulty in ensuring the stability and accuracy of the vehicle track.

[0037] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0038] Example 1

[0039] Figure 1 A schematic diagram of the process of updating the vehicle track provided in this application Figure 1 ,like Figure 1 As shown, the method includes:

[0040] S101, obtaining a visual obstacle and multiple radar obstacles corresponding to the vehicle, screening the multiple radar obstacles to obtain multiple first candidate obstacles;

[0041] S102, determining an association probability between each first candidate obstacle and the visual obstacle based on the attribute information of the plurality of first candidate obstacles and the attribute information of the visual obstacle, and determining a plurality of second candidate obstacles based on the association probability;

[0042] S103, determining a first distance parameter of each second candidate obstacle, and determining a target obstacle from the plurality of second candidate obstacles based on the first distance parameter;

[0043] S104: Update the vehicle's track based on the target obstacle.

[0044] The target obstacle is the optimal radar obstacle for updating the vehicle's track, and the first and second candidate obstacles are candidates for the optimal radar obstacle. Furthermore, the distance parameter in the first distance parameter of each second candidate obstacle may be a Mahalanobis distance. The target obstacle is selected from the candidate obstacles.

[0045] In practical applications, obtain the visual obstacles and multiple radar obstacles corresponding to the vehicle, Figure 2 A schematic diagram of a radar obstacle, such as Figure 2 As shown in the figure, taking the car as an example, the visual obstacle is the rear of the observed vehicle. For this car, the radar provides four radar obstacles corresponding to the car. The four radar obstacles are radar 1, radar 2, radar 3, and radar 4. The four radar obstacles are screened to obtain multiple first candidate obstacles, namely radar 1, radar 2, and radar 3. The association probability between each first candidate obstacle and the visual obstacle is determined based on the attribute information of the multiple first candidate obstacles and the attribute information of the visual obstacle. For example, the association probability between radar 1 and the visual obstacle is determined based on the attribute information of radar 1 and the attribute information of the visual obstacle. In this way, multiple second candidate obstacles are obtained. The first Mahalanobis distance of each second candidate obstacle is determined, and based on the first Mahalanobis distance, the target obstacle is determined from the multiple second candidate obstacles to update the vehicle's track.

[0046] Specifically, a vehicle's track refers to the trajectory formed by the vehicle's position changing over time during driving; it is a dynamic record and representation of the vehicle's driving path; updating the vehicle's track based on target obstacles can optimize the driving path and enhance driving safety.

[0047] The system obtains visual obstacles and multiple radar obstacles corresponding to the vehicle, screens the multiple radar obstacles, and obtains multiple first obstacle candidates. Based on the attribute information of the multiple first obstacle candidates and the attribute information of the visual obstacle, it determines the association probability between each first obstacle candidate and the visual obstacle. Based on the association probability, it screens and determines multiple second obstacle candidates. For each of the obtained second obstacle candidates, it calculates a first distance parameter and, based on the first distance parameter, determines a target obstacle from the multiple second candidate obstacles. Based on the target obstacle, it updates the vehicle's track. This solution updates the vehicle's track by associating visual obstacles with radar obstacles and calculating the first distance parameter to select the target obstacle, thereby improving the stability and accuracy of the vehicle's track.

[0048] In some examples, the vehicle's trajectory is updated based on the target obstacle, including:

[0049] Determine whether the target obstacle is the currently associated target obstacle;

[0050] If the target obstacle is not the currently associated target obstacle, the target obstacle is associated with the vehicle's track to obtain a new track, and the association between the new track and the visual obstacle is updated; the attributes of the new track are filtered and smoothed; and the vehicle's track is updated to the new track;

[0051] If the target obstacle is the currently associated target obstacle, the target obstacle is associated with the vehicle's track to obtain a new track, the association relationship between the new track and the visual obstacle is updated; and the vehicle's track is updated to the new track.

[0052] Filtering and smoothing the attributes of the new track refers to applying filtering and smoothing techniques to process the newly generated attribute data related to the vehicle's track. Specifically, when acquiring data, the vehicle's sensors inevitably introduce noise. For example, radar signals may be subject to electromagnetic interference, and Global Positioning System (GPS) signals may produce errors due to factors such as the atmosphere. This noise can cause fluctuations and inaccuracies in the track data. Filtering and smoothing can effectively remove this noise, allowing the track to more accurately reflect the vehicle's actual driving path. Furthermore, filtering and smoothing can reduce data fluctuations, allowing track attribute data, such as position, velocity, and acceleration, to change more smoothly between adjacent moments.

[0053] In practical applications, when the target obstacle to be associated changes, the data obtained by the sensor about the new target obstacle will differ from the original target obstacle data, resulting in a sudden change in the track data. Therefore, the properties of the new track obtained after association need to be filtered and smoothed. If the target obstacle to be associated does not change, the sensor continues to monitor the same target obstacle, and the data obtained has high consistency and stability, and the track data does not produce significant fluctuations or noise. Therefore, no additional filtering and smoothing is required to adjust. Filtering and smoothing consumes computing resources and time, and performing filtering and smoothing even when the target obstacle to be associated has not changed is a waste of resources.

[0054] Specifically, when updating the vehicle's track, first determine whether the target obstacle is the currently associated target obstacle; then perform different update processing based on the judgment result; if the target obstacle is not the currently associated target obstacle, then associate the target obstacle with the vehicle's track to obtain a new track, and update the association relationship between the new track and the visual obstacle; and filter and smooth the attributes of the new track; then update the vehicle's track to the new track; if the target obstacle is the currently associated target obstacle, then associate the target obstacle with the vehicle's track to obtain a new track, and update the association relationship between the new track and the visual obstacle; and then update the vehicle's track to the new track. Filtering and smoothing the attributes of the new track obtained based on the changed target obstacle association can effectively ensure the smoothness of the track attributes, reduce data jitter and errors, maintain the stability of the track, ensure the continuity and reliability of the track, and provide more accurate and reliable data support for subsequent tracking and prediction.

[0055] In some examples, a plurality of radar obstacles are screened to obtain a plurality of first candidate obstacles, including:

[0056] determining, for any one of the multiple radar obstacles, a longitudinal distance error and a lateral distance error between the radar obstacle and the visual obstacle, and determining a distance error value corresponding to the radar obstacle based on the longitudinal distance error and the lateral distance error;

[0057] Determine whether the distance error value is less than the preset error value;

[0058] If the distance error value is less than the preset error value, the radar obstacle is taken as the first candidate obstacle;

[0059] If the distance error value is not less than the preset error value, the radar obstacle is removed.

[0060] The preset error value is set as needed. For example, the preset error value is set to 1. In conjunction with the above example, for any one of the multiple radar obstacles, the longitudinal distance error and lateral distance error between the radar obstacle and the visual obstacle are determined. For example, for Radar 1, Radar 2, and Radar 3, the longitudinal distance errors and lateral distance errors between Radar 1, Radar 2, and Radar 3 and the visual obstacle are determined, respectively. Based on the longitudinal distance errors and lateral distance errors, the distance error values ​​corresponding to the different radar obstacles are calculated. All calculated different radar obstacles are used to determine whether their distance error values ​​are less than the preset error value. Specifically, for example, if the distance error values ​​corresponding to Radar 1 and Radar 2 are both less than 1, Radar 1 and Radar 2 are the first candidate obstacles. If the distance error value corresponding to Radar 3 is greater than 1, Radar 3 is eliminated.

[0061] By calculating the distance error value of radar obstacles and then screening radar obstacles, it is possible to avoid misjudging radar obstacles as candidate obstacles and selecting them as target obstacles, which will cause vehicle tracking errors. At the same time, it also enables the radar system to maintain good obstacle detection and tracking capabilities in complex environments.

[0062] In some examples, determining a distance error value corresponding to a radar obstacle based on the longitudinal distance error and the lateral distance error includes:

[0063] The longitudinal distance error is divided by the longitudinal preset parameter and then squared to obtain the longitudinal distance error value; the lateral distance error is divided by the lateral preset parameter and then squared to obtain the lateral distance error value;

[0064] The longitudinal distance error value and the lateral distance error value are summed to obtain the distance error value corresponding to the radar obstacle.

[0065] The preset parameters can be set according to needs, for example, the longitudinal preset parameter is set to 1.5, and the transverse preset parameter is set to 1.8.

[0066] In practical applications, the distance error value corresponding to the radar obstacle is determined based on the longitudinal distance error and the lateral distance error. Combining the previous example, the following formula can be used:

[0067]

[0068] in, For vertical preset parameters, For horizontal preset parameters, is the longitudinal distance error between the visual target and the candidate radar, is the lateral distance error between the visual target and the candidate radar, and 1 is the preset error value.

[0069] Specifically, for example, the error value of radar 1 is calculated; the longitudinal distance error and the lateral distance error between the radar obstacle and the visual obstacle are determined to be Mihe m, substitute all the values ​​into the formula on the left to get , we can see that the error value of radar 1, 0.89, is less than 1.

[0070] This formula can be used to calculate the longitudinal and lateral position errors of the vehicle. 、 Perform comprehensive quantization. By dividing the error by and The squares are then added together to produce a single numerical result. For example, in autonomous driving scenarios, the system can use this value to accurately determine the degree of deviation between the vehicle's current position and the expected position, rather than considering the errors in two directions in isolation. This helps to more comprehensively evaluate the accuracy of vehicle positioning.

[0071] In some examples, determining an association probability between each first candidate obstacle and the visual obstacle based on attribute information of the first candidate obstacle and attribute information of the visual obstacle, and determining a plurality of second candidate obstacles based on the association probabilities includes:

[0072] For any one of the plurality of first candidate obstacles, determining whether the number of attribute information of the first candidate obstacle and the attribute information of the visual obstacle that are consistent reaches a preset number;

[0073] If the attribute consistency reaches the preset number of items, the association probability of the first candidate obstacle is not zero, and the first candidate obstacle with a non-zero association probability is used as the second candidate obstacle;

[0074] If the attribute consistency does not reach the preset number of items, the association probability of the first candidate obstacle is zero.

[0075] The attribute information of the first obstacle candidate and the visual obstacle includes, but is not limited to, motion state, direction, and historical trajectory. These three attributes are used to compare the visual obstacle with each first obstacle candidate. Specifically, regarding motion state, the obstacle's motion state is categorized as either stationary or moving. If the visual obstacle appears stationary while the first obstacle candidate is in motion, the two are considered to have significantly different motion state attributes. Regarding direction, the visual obstacle and the first obstacle candidate each have different directions if the difference exceeds a preset range. Regarding historical trajectory, both the visual obstacle and the first obstacle candidate store multiple frames of historical global coordinates. These coordinates enable accurate determination of whether the target is moving forward, backward, left, or right, or stationary. If there are significant differences, the historical trajectories are considered inconsistent. After evaluating the three attributes, if two or more attributes match, the probability of association with the first obstacle candidate is non-zero, and the first obstacle candidate with a non-zero association probability is selected as the second obstacle candidate. If only one or no attributes match, the probability of association with the first obstacle candidate is non-zero.

[0076] In practical applications, for any of the multiple first candidate obstacles, a determination is made as to whether the attribute information of the first candidate obstacle matches a preset number of attribute matches between the first candidate obstacle and the visual obstacle. If the attribute matches, the association probability of the first candidate obstacle is non-zero, and the first candidate obstacle with a non-zero association probability is selected as the second candidate obstacle. If the attribute matches do not match, the association probability of the first candidate obstacle is zero. For example, if there are 10 first candidate obstacles, after comparing the attributes, seven first candidate obstacles with non-zero association probabilities are selected as the second candidate obstacles, and three first candidate obstacles with zero association probabilities are selected. Specifically, if the association probabilities of all ten first candidate obstacles are zero after the attribute comparison, the track established with the visual obstacle is temporarily used, and the process returns to the step of selecting multiple first radar obstacles from the multiple radar obstacles.

[0077] By comparing the attribute information, some impossible first candidate obstacles can be quickly filtered out, and the first candidate obstacle and the visual obstacle are not the same obstacle, which leads to track attribute update errors caused by forced association errors.

[0078] In some examples, Figure 3 A schematic diagram of the process of updating the vehicle track provided in this application Figure 2 ,like Figure 3 As shown, determining the first distance parameter of each second candidate obstacle includes:

[0079] S301: For any one of the plurality of second candidate obstacles and the visual obstacle, determine a state matrix of the visual obstacle and a state matrix of the second candidate obstacle, and determine a second distance parameter of the second candidate obstacle based on the state matrix of the visual obstacle and the state matrix of the second candidate obstacle; wherein the state matrix represents position data and velocity data of the second candidate obstacle and the visual obstacle;

[0080] S302, adjusting preset weights of different directions based on historical experience to obtain a first weight; wherein the different directions include longitudinal position, transverse position, longitudinal speed, and transverse speed;

[0081] S303, performing calculation based on the second distance parameter and the first weight to obtain a third distance parameter;

[0082] S304: Obtain the number of times the second candidate obstacle is associated, and calculate the first distance parameter using a nonlinear activation function based on the third distance parameter and the number of times the second candidate obstacle is associated;

[0083] S305: Filter out the minimum distance parameter from the first distance parameters.

[0084] Among them, the distance parameter can be the Mahalanobis distance; the state quantity matrix is ​​as follows:

[0085]

[0086] in, Indicates the vertical position, Indicates horizontal position, Indicates vertical speed, For example, for any second candidate obstacle, the state matrix of the second candidate obstacle is ; The state matrix of the visual obstacle is .

[0087] Additionally, the preset weights for different directions are as follows:

[0088]

[0089] For example, the preset initial weights are , , , Adjust the preset weights of different directions based on historical experience. For example, if the current track is judged to be in a close-range straight-ahead state based on historical experience, adjust the weights. , , then the first weight is ; If the track is judged to be in a long-distance curve, adjust the weight , then the first weight is ; If the track is judged to be in a cross-state, adjust the weight to , then the first weight is .

[0090] In practical applications, the second Mahalanobis distance of the second candidate obstacle is determined based on the state matrix of the visual obstacle and the state matrix of the second candidate obstacle. Based on historical experience, the preset weights for different directions are adjusted to obtain the first weight. The third Mahalanobis distance is calculated based on the second Mahalanobis distance and the first weight. The number of times the second candidate obstacle is associated is obtained, and the first Mahalanobis distance is calculated using a nonlinear activation function based on the third Mahalanobis distance and the number of times the second candidate obstacle is associated. The smallest Mahalanobis distance is selected from the first Mahalanobis distances. By adjusting the weights to calculate the Mahalanobis distance, the vehicle can quickly adapt to environmental changes and quickly find the optimal driving path in complex environments.

[0091] In some examples, determining a second distance parameter of the second candidate obstacle based on the state matrix of the visual obstacle and the state matrix of the second candidate obstacle includes:

[0092] Subtracting the state matrix of the visual obstacle from the state matrix of the second candidate obstacle, and transposing the resulting matrix to obtain a first transposed matrix of the visual obstacle and the second candidate obstacle;

[0093] Obtain an inverse matrix of a first covariance matrix between the visual obstacle and the second candidate obstacle according to the first transposed matrix;

[0094] The second distance parameter is obtained by multiplying the first transposed matrix by the inverse matrix of the first covariance matrix.

[0095] Combined with the previous example, the formula for calculating the second distance parameter is the Mahalanobis distance formula, as shown below:

[0096]

[0097] in, represents the Mahalanobis distance, Represents the transposed matrix of visual obstacles and candidate obstacles, represents the inverse matrix of the covariance matrix between visual obstacles and candidate obstacles, The covariance matrix between visual obstacles and candidate obstacles is shown as follows:

[0098]

[0099] Specifically, combined with the previous example, the state matrix of any second candidate obstacle is ; The state matrix of visual obstacles , calculated In practical applications, the covariance needs to be determined according to the specific situation. To facilitate understanding and calculation, it is assumed here that the speed and position variables are independent of each other. However, in actual data, there is correlation between the variables and they are not independent of each other. Then the covariance matrix ∑ is:

[0100] but , , =10.5. Calculating the distance parameter using the Mahalanobis distance formula can take into account the correlation between variables and improve the accuracy of the calculation results.

[0101] In some examples, calculating the third distance parameter based on the second distance parameter and the first weight includes:

[0102] After subtracting the state matrix of the visual obstacle from the state matrix of the second candidate obstacle, the obtained matrix is ​​multiplied by the first weight and then transposed to obtain a second transposed matrix of the visual obstacle and the second candidate obstacle;

[0103] Obtain an inverse matrix of a second covariance matrix between the visual obstacle and the second candidate obstacle according to the second transposed matrix;

[0104] The third distance parameter is obtained by multiplying the second transposed matrix by the inverse matrix of the second covariance matrix.

[0105] Combining the above example, the formula for calculating the third distance parameter is as follows:

[0106]

[0107] The adjusted weight is In the case of , the calculation result of the third distance parameter is .

[0108] By improving the Mahalanobis distance formula, adding weights to the original Mahalanobis distance formula, and adjusting the weights according to different scenarios, the Mahalanobis distance calculation results can be made more in line with the actual operation needs of the vehicle.

[0109] In some examples, the first distance parameter is obtained by calculating the third distance parameter and the number of times the second candidate obstacle is associated using a nonlinear activation function, including:

[0110] The number of times the second candidate obstacle is associated is subtracted and divided by a preset parameter. The result is calculated using a nonlinear activation function to obtain an optimized value.

[0111] The first distance parameter is obtained by multiplying the result obtained by subtracting the preset value from the optimized value by the third distance parameter.

[0112] Specifically, considering that the longer a track is associated with a fixed obstacle, the greater the probability that they are the same obstacle, it is necessary to "encourage" them to continue to be associated. Therefore, the formula for calculating the first distance parameter is as follows:

[0113]

[0114] in, represents a nonlinear activation function, Indicates the number of times the second candidate obstacle is associated, Indicates the preset parameters. represents the third distance parameter calculated in the previous example. Represents the first distance parameter. Specifically, the nonlinear activation function is: ; In the function In combination with the previous example, for example, the second candidate obstacle is associated 4 times, and the preset parameter is 1.6, then 2.5, substituting it into the nonlinear activation function, we can get ; .

[0115] In practice, the optimization value is calculated using a nonlinear activation function, which is calculated by subtracting the number of times the second candidate obstacle has been associated and dividing it by a preset parameter. The result is then subtracted from the preset value and multiplied by the third distance parameter to obtain the first distance parameter. As the association binding time increases, the value of the distance parameter decreases. At the same time, the inherent characteristics of the Sigmoid function ensure that when bind_cnt is small, the "encouragement" of association behavior is not obvious. As bind_cnt gradually increases, the compensation increases. Even when bind_cnt is infinite, the Mahalanobis distance score will not be negative.

[0116] In some examples, the method further includes:

[0117] Determine whether the third distance parameter corresponding to the second candidate obstacle corresponding to the minimum distance parameter is a minimum value;

[0118] If the corresponding third distance parameter is not the minimum value, the number of times the second candidate obstacle corresponding to the minimum distance parameter is associated is updated according to the exponential function, the first distance parameter is recalculated and screened, and the second candidate obstacle corresponding to the minimum distance parameter obtained through screening is used as the target obstacle;

[0119] If the corresponding third distance parameter is the minimum value, the second candidate obstacle corresponding to the minimum distance parameter is used as the target obstacle.

[0120] To avoid the situation where the track is associated with the current target obstacle while the target obstacle is not present and compensation is continuously being applied to the current target obstacle, even if the target obstacle subsequently appears, the calculated first distance parameter will be greater than the compensated first distance parameter of the current target obstacle, making association impossible. Therefore, before determining the target obstacle, the calculated third distance parameter of each candidate obstacle before compensation is constantly compared.

[0121] In practical applications, combined with the previous example, the smallest distance parameter among the first distance parameters calculated is , and the corresponding third distance parameter of the second candidate obstacle is , the third distance parameter Compare the third distance parameter with the remaining third distance parameters. If the third distance parameter is not the minimum value, and the minimum value of the third distance parameter is 16.35, then use the exponential function to update the number of associations for the second candidate obstacle with a third distance parameter of 16.75. Recalculate the first distance parameter using the distance parameter formula and perform screening. The second candidate obstacle corresponding to the minimum distance parameter obtained in this screening is used as the target obstacle. If the third distance parameter is the minimum value, then the second candidate obstacle corresponding to the minimum distance parameter is used as the target obstacle.

[0122] In some examples, updating the number of times the second candidate obstacle is associated corresponding to the minimum distance parameter according to an exponential function includes:

[0123] The number of association errors of the second candidate obstacle corresponding to the minimum distance parameter is increased by one and then the exponent is taken. The result is multiplied by the number of times the second candidate obstacle is associated and rounded down to obtain the new number of times the second candidate obstacle is associated.

[0124] Specifically, the formula for updating the number of times the second candidate obstacle is associated corresponding to the minimum distance parameter is: ,in, Indicates the number of association errors; for example, combined with the above example, assuming that the current number of association errors is 1, add 1 to the current number of association errors to get , substituting into the formula, the new number of associations of the second candidate obstacle is By updating the number of times the second candidate obstacle is associated, the value of bind_cnt can be quickly reduced, which is conducive to switching from the wrong associated target to the correct associated target in time.

[0125] In some examples, after the association probability of the radar of the first candidate obstacle is zero, the method further includes:

[0126] The first obstacle candidate with an association probability of zero is used as the third obstacle candidate, and the third obstacle candidate is returned to execute the above steps starting from the screening process of the plurality of radar obstacles.

[0127] Specifically, when the association probability of the radar of the first candidate obstacle is zero, it means that under the current situation, the first candidate obstacle does not meet the requirements; the first candidate obstacle is selected as the third candidate obstacle and the steps of screening multiple radar obstacles in this solution are performed again. Figure 4 Select a schematic diagram for a target obstacle, such as Figure 4 As shown in the figure, when the distance is far, the visual obstacle attributes have not converged and fluctuate significantly. First, the target obstacle (Radar 3) with the closest attributes is selected for association to ensure the track appears as early as possible, which is beneficial for vehicle control. During the screening process, if Radar 3's attributes are determined to be non-compliant, the association probability is zero, and Radar 1, the target obstacle, is selected from the remaining candidate obstacles for association. Re-including the first candidate obstacle with a zero association probability as the third candidate in the screening process can prevent any association misjudgments caused by data anomalies in the early stages, as they can be corrected in the next round of screening. This makes the data processing process more robust, reduces the impact of erroneous data on the final results, and improves the system's ability to cope with complex situations.

[0128] An embodiment of the present application provides a vehicle track updating method. The method comprises the following steps: first, obtaining a visual obstacle and multiple radar obstacles corresponding to the vehicle, screening the multiple radar obstacles to obtain multiple first obstacle candidates; determining an association probability between each first obstacle candidate and the visual obstacle based on attribute information of the multiple first obstacle candidates and attribute information of the visual obstacle; and screening and determining multiple second obstacle candidates based on the association probability; calculating a first distance parameter for each of the obtained second obstacle candidates, and determining a target obstacle from the multiple second obstacle candidates based on the first distance parameter; and updating the vehicle track based on the target obstacle.

[0129] Combined with the above examples, in practical applications, Figure 5 Schematic diagram of the process of a vehicle track update method provided by this application Figure 3 ,like Figure 5As shown; a visual obstacle and multiple radar obstacles are obtained, the multiple radar obstacles are screened to obtain multiple first candidate obstacles, and the motion state, direction, and historical trajectory of the visual obstacle and each first candidate obstacle are compared; the first candidate obstacle with an association probability of zero is returned as a third candidate obstacle to execute the step of obtaining multiple first candidate obstacles, the first candidate obstacle with an association probability not zero is used as a second candidate obstacle, and the first distance parameter of each second candidate obstacle is calculated, the second candidate obstacle corresponding to the smallest distance parameter is selected as the target obstacle, and a new track is obtained based on the target obstacle; it is determined whether the target obstacle is the currently associated target obstacle. If so, the vehicle track is updated to the new track; if not, the attributes of the new track are filtered and smoothed, and the vehicle track is updated to the new track.

[0130] Example 2

[0131] Figure 6 This is a schematic diagram of the structure of a vehicle track updating device provided by this application, such as Figure 6 As shown, including:

[0132] an acquisition module 10 for acquiring a visual obstacle and a plurality of radar obstacles corresponding to the vehicle, and screening the plurality of radar obstacles to obtain a plurality of first candidate obstacles;

[0133] a processing module 20 for determining an association probability between each first candidate obstacle and the visual obstacle based on the attribute information of the plurality of first candidate obstacles and the attribute information of the visual obstacle, and determining a plurality of second candidate obstacles based on the association probability;

[0134] The processing module 20 is further configured to determine a first distance parameter of each second candidate obstacle, and determine a target obstacle from the plurality of second candidate obstacles based on the first distance parameter;

[0135] The processing module 20 is further configured to update the vehicle's track based on the target obstacle.

[0136] The target obstacle is the optimal radar obstacle for updating the vehicle's track, and the first and second candidate obstacles are candidates for the optimal radar obstacle. Furthermore, the distance parameter in the first distance parameter of each second candidate obstacle may be a Mahalanobis distance. The target obstacle is selected from the candidate obstacles.

[0137] In practical applications, visual obstacles and multiple radar obstacles corresponding to the vehicle are obtained. For example, a car is observed as the visual obstacle at the rear of the vehicle. For this car, the radar generates four radar obstacles corresponding to the car. The four radar obstacles are radar 1, radar 2, radar 3, and radar 4. The four radar obstacles are screened to obtain multiple first candidate obstacles, namely radar 1, radar 2, and radar 3. The association probability between each first candidate obstacle and the visual obstacle is determined based on the attribute information of the multiple first candidate obstacles and the attribute information of the visual obstacle. For example, the association probability between radar 1 and the visual obstacle is determined based on the attribute information of radar 1 and the attribute information of the visual obstacle. This results in multiple second candidate obstacles. The first Mahalanobis distance of each second candidate obstacle is determined, and based on this first Mahalanobis distance, the target obstacle is identified from the multiple second candidate obstacles to update the vehicle's track.

[0138] Specifically, a vehicle's track refers to the trajectory formed by the vehicle's position changing over time during driving; it is a dynamic record and representation of the vehicle's driving path; updating the vehicle's track based on target obstacles can optimize the driving path and enhance driving safety.

[0139] The system obtains visual obstacles and multiple radar obstacles corresponding to the vehicle, screens the multiple radar obstacles, and obtains multiple first obstacle candidates. The system then determines the association probability between each first obstacle candidate and the visual obstacle based on the attribute information of the multiple first obstacle candidates and the attribute information of the visual obstacle. Based on the association probability, multiple second obstacle candidates are screened and determined. For each of the obtained second obstacle candidates, a first distance parameter is calculated, and based on the first distance parameter, a target obstacle is determined from the multiple second candidate obstacles. The vehicle's track is then updated based on the target obstacle. This solution updates the vehicle's track by associating visual obstacles with radar obstacles and calculating the first distance parameter to screen out the target obstacle, ensuring the stability of the vehicle's track.

[0140] In some examples, the vehicle's trajectory is updated based on the target obstacle, and the processing module 20 is further configured to:

[0141] Determine whether the target obstacle is the currently associated target obstacle;

[0142] If the target obstacle is not the currently associated target obstacle, the target obstacle is associated with the vehicle's track to obtain a new track, and the association between the new track and the visual obstacle is updated; the attributes of the new track are filtered and smoothed; and the vehicle's track is updated to the new track;

[0143] If the target obstacle is the currently associated target obstacle, the target obstacle is associated with the vehicle's track to obtain a new track, the association relationship between the new track and the visual obstacle is updated; and the vehicle's track is updated to the new track.

[0144] Filtering and smoothing the attributes of the new track refers to applying filtering and smoothing techniques to process the newly generated attribute data related to the vehicle's track. Specifically, when acquiring data, the vehicle's sensors inevitably introduce noise. For example, radar signals may be subject to electromagnetic interference, and Global Positioning System (GPS) signals may produce errors due to factors such as the atmosphere. This noise can cause fluctuations and inaccuracies in the track data. Filtering and smoothing can effectively remove this noise, allowing the track to more accurately reflect the vehicle's actual driving path. Furthermore, filtering and smoothing can reduce data fluctuations, allowing track attribute data, such as position, velocity, and acceleration, to change more smoothly between adjacent moments.

[0145] In practical applications, when the target obstacle to be associated changes, the data obtained by the sensor about the new target obstacle will differ from the original target obstacle data, resulting in a sudden change in the track data. Therefore, the properties of the new track obtained after association need to be filtered and smoothed. If the target obstacle to be associated does not change, the sensor continues to monitor the same target obstacle, and the data obtained has high consistency and stability, and the track data does not produce significant fluctuations or noise. Therefore, no additional filtering and smoothing is required to adjust. Filtering and smoothing consumes computing resources and time, and performing filtering and smoothing even when the target obstacle to be associated has not changed is a waste of resources.

[0146] Specifically, when updating the vehicle's track, first determine whether the target obstacle is the currently associated target obstacle; then perform different update processing based on the judgment result; if the target obstacle is not the currently associated target obstacle, then associate the target obstacle with the vehicle's track to obtain a new track, and update the association relationship between the new track and the visual obstacle; and filter and smooth the attributes of the new track; then update the vehicle's track to the new track; if the target obstacle is the currently associated target obstacle, then associate the target obstacle with the vehicle's track to obtain a new track, and update the association relationship between the new track and the visual obstacle; and then update the vehicle's track to the new track. Filtering and smoothing the attributes of the new track obtained based on the changed target obstacle association can effectively ensure the smoothness of the track attributes, reduce data jitter and errors, maintain the stability of the track, ensure the continuity and reliability of the track, and provide more accurate and reliable data support for subsequent tracking and prediction.

[0147] In some examples, after screening multiple radar obstacles to obtain multiple first candidate obstacles, the acquisition module 10 is further configured to:

[0148] determining, for any one of the multiple radar obstacles, a longitudinal distance error and a lateral distance error between the radar obstacle and the visual obstacle, and determining a distance error value corresponding to the radar obstacle based on the longitudinal distance error and the lateral distance error;

[0149] Determine whether the distance error value is less than the preset error value;

[0150] If the distance error value is less than the preset error value, the radar obstacle is taken as the first candidate obstacle;

[0151] If the distance error value is not less than the preset error value, the radar obstacle is removed.

[0152] The preset error value is set as needed. For example, the preset error value is set to 1. In conjunction with the above example, for any one of the multiple radar obstacles, the longitudinal distance error and lateral distance error between the radar obstacle and the visual obstacle are determined. For example, for Radar 1, Radar 2, and Radar 3, the longitudinal distance errors and lateral distance errors between Radar 1, Radar 2, and Radar 3 and the visual obstacle are determined, respectively. Based on the longitudinal distance errors and lateral distance errors, the distance error values ​​corresponding to the different radar obstacles are calculated. All calculated different radar obstacles are used to determine whether their distance error values ​​are less than the preset error value. Specifically, for example, if the distance error values ​​corresponding to Radar 1 and Radar 2 are both less than 1, Radar 1 and Radar 2 are the first candidate obstacles. If the distance error value corresponding to Radar 3 is greater than 1, Radar 3 is eliminated.

[0153] By calculating the distance error value of radar obstacles and then screening radar obstacles, it is possible to avoid misjudging radar obstacles as candidate obstacles and selecting them as target obstacles, which will cause vehicle tracking errors. At the same time, it also enables the radar system to maintain good obstacle detection and tracking capabilities in complex environments.

[0154] In some examples, based on the longitudinal distance error and the lateral distance error, the distance error value corresponding to the radar obstacle is determined, and the processing module 20 is further configured to:

[0155] The longitudinal distance error is divided by the longitudinal preset parameter and then squared to obtain the longitudinal distance error value; the lateral distance error is divided by the lateral preset parameter and then squared to obtain the lateral distance error value;

[0156] The longitudinal distance error value and the lateral distance error value are summed to obtain the distance error value corresponding to the radar obstacle.

[0157] The preset parameters can be set according to needs, for example, the longitudinal preset parameter is set to 1.5, and the transverse preset parameter is set to 1.8.

[0158] In practical applications, the distance error value corresponding to the radar obstacle is determined based on the longitudinal distance error and the lateral distance error. Combining the previous example, the following formula can be used:

[0159]

[0160] in, For vertical preset parameters, For horizontal preset parameters, is the longitudinal distance error between the visual target and the candidate radar, is the lateral distance error between the visual target and the candidate radar, and 1 is the preset error value.

[0161] Specifically, for example, the error value of radar 1 is calculated; the longitudinal distance error and the lateral distance error between the radar obstacle and the visual obstacle are determined to be Mihe m, substitute all the values ​​into the formula on the left to get , we can see that the error value of radar 1, 0.89, is less than 1.

[0162] This formula can be used to calculate the longitudinal and lateral position errors of the vehicle. 、 Perform comprehensive quantization. By dividing the error by and The squares are then added together to produce a single numerical result. For example, in autonomous driving scenarios, the system can use this value to accurately determine the degree of deviation between the vehicle's current position and the expected position, rather than considering the errors in two directions in isolation. This helps to more comprehensively evaluate the accuracy of vehicle positioning.

[0163] In some examples, the processing module 20 is further configured to determine an association probability between each first candidate obstacle and the visual obstacle based on the attribute information of the first candidate obstacle and the attribute information of the visual obstacle, and determine multiple second candidate obstacles based on the association probability:

[0164] For any one of the plurality of first candidate obstacles, determining whether the number of attribute information of the first candidate obstacle and the attribute information of the visual obstacle that are consistent reaches a preset number;

[0165] If the attribute consistency reaches the preset number of items, the association probability of the first candidate obstacle is not zero, and the first candidate obstacle with a non-zero association probability is used as the second candidate obstacle;

[0166] If the attribute consistency does not reach the preset number of items, the association probability of the first candidate obstacle is zero.

[0167] The attribute information of the first obstacle candidate and the visual obstacle includes, but is not limited to, motion state, direction, and historical trajectory. These three attributes are used to compare the visual obstacle with each first obstacle candidate. Specifically, regarding motion state, the obstacle's motion state is categorized as either stationary or moving. If the visual obstacle appears stationary while the first obstacle candidate is in motion, the two are considered to have significantly different motion state attributes. Regarding direction, the visual obstacle and the first obstacle candidate each have different directions if the difference exceeds a preset range. Regarding historical trajectory, both the visual obstacle and the first obstacle candidate store multiple frames of historical global coordinates. These coordinates enable accurate determination of whether the target is moving forward, backward, left, or right, or stationary. If there are significant differences, the historical trajectories are considered inconsistent. After evaluating the three attributes, if two or more attributes match, the probability of association with the first obstacle candidate is non-zero, and the first obstacle candidate with a non-zero association probability is selected as the second obstacle candidate. If only one or no attributes match, the probability of association with the first obstacle candidate is non-zero.

[0168] In practical applications, for any of the multiple first candidate obstacles, a determination is made as to whether the attribute information of the first candidate obstacle matches a preset number of attribute matches between the first candidate obstacle and the visual obstacle. If the attribute matches, the association probability of the first candidate obstacle is non-zero, and the first candidate obstacle with a non-zero association probability is selected as the second candidate obstacle. If the attribute matches do not match, the association probability of the first candidate obstacle is zero. For example, if there are 10 first candidate obstacles, after comparing the attributes, seven first candidate obstacles with non-zero association probabilities are selected as the second candidate obstacles, and three first candidate obstacles with zero association probabilities are selected. Specifically, if the association probabilities of all ten first candidate obstacles are zero after the attribute comparison, the track established with the visual obstacle is temporarily used, and the process returns to the step of selecting multiple first radar obstacles from the multiple radar obstacles.

[0169] By comparing the attribute information, some impossible first candidate obstacles can be quickly filtered out, and the first candidate obstacle and the visual obstacle are not the same obstacle, which leads to track attribute update errors caused by forced association errors.

[0170] In some examples, after determining the first distance parameter of each second candidate obstacle, the processing module 20 is further configured to:

[0171] For any one of the plurality of second candidate obstacles and the visual obstacle, determining a state quantity matrix of the visual obstacle and a state quantity matrix of the second candidate obstacle, and determining a second distance parameter of the second candidate obstacle based on the state quantity matrix of the visual obstacle and the state quantity matrix of the second candidate obstacle; wherein the state quantities represent position data and velocity data of the second candidate obstacle and the visual obstacle;

[0172] Adjusting the preset weights of different directions based on historical experience to obtain a first weight; wherein the different directions include longitudinal position, transverse position, longitudinal speed, and transverse speed;

[0173] Calculating according to the second distance parameter and the first weight to obtain a third distance parameter;

[0174] Obtain the number of times the second candidate obstacle is associated, and calculate the first distance parameter using a nonlinear activation function based on the third distance parameter and the number of times the second candidate obstacle is associated;

[0175] The smallest distance parameter is selected from the first distance parameters.

[0176] Among them, the distance parameter can be the Mahalanobis distance; the state quantity matrix is ​​as follows:

[0177]

[0178] in, Indicates the vertical position, Indicates horizontal position, Indicates vertical speed, For example, for any second candidate obstacle, the state matrix of the second candidate obstacle is ; The state matrix of the visual obstacle is .

[0179] Additionally, the preset weights for different directions are as follows:

[0180]

[0181] For example, the preset initial weights are , , , Adjust the preset weights of different directions based on historical experience. For example, if the current track is judged to be in a close-range straight-ahead state based on historical experience, adjust the weights. , , then the first weight is ; If the track is judged to be in a long-distance curve, adjust the weight , then the first weight is ; If the track is judged to be in a cross-state, adjust the weight to , then the first weight is .

[0182] In practical applications, the second Mahalanobis distance of the second candidate obstacle is determined based on the state matrix of the visual obstacle and the state matrix of the second candidate obstacle. Based on historical experience, the preset weights for different directions are adjusted to obtain the first weight. The third Mahalanobis distance is calculated based on the second Mahalanobis distance and the first weight. The number of times the second candidate obstacle is associated is obtained, and the first Mahalanobis distance is calculated using a nonlinear activation function based on the third Mahalanobis distance and the number of times the second candidate obstacle is associated. The smallest Mahalanobis distance is selected from the first Mahalanobis distances. By adjusting the weights to calculate the Mahalanobis distance, the vehicle can quickly adapt to environmental changes and quickly find the optimal driving path in complex environments.

[0183] In some examples, based on the state matrix of the visual obstacle and the state matrix of the second candidate obstacle, the second distance parameter of the second candidate obstacle is determined, and the processing module 20 is further configured to:

[0184] Subtracting the state matrix of the visual obstacle from the state matrix of the second candidate obstacle, and transposing the resulting matrix to obtain a first transposed matrix of the visual obstacle and the second candidate obstacle;

[0185] Obtain an inverse matrix of a first covariance matrix between the visual obstacle and the second candidate obstacle according to the first transposed matrix;

[0186] The second distance parameter is obtained by multiplying the first transposed matrix by the inverse matrix of the first covariance matrix.

[0187] Combined with the previous example, the formula for calculating the second distance parameter is the Mahalanobis distance formula, as shown below:

[0188]

[0189] in, represents the Mahalanobis distance, Represents the transposed matrix of visual obstacles and candidate obstacles, represents the inverse matrix of the covariance matrix between visual obstacles and candidate obstacles, The covariance matrix between visual obstacles and candidate obstacles is shown as follows:

[0190]

[0191] Specifically, combined with the previous example, the state matrix of any second candidate obstacle is ; The state matrix of visual obstacles , calculated In practical applications, the covariance needs to be determined according to the specific situation. To facilitate understanding and calculation, it is assumed here that the speed and position variables are independent of each other. However, in actual data, there is correlation between the variables and they are not independent of each other. Then the covariance matrix ∑ is:

[0192] but , , =10.5. Calculating the distance parameter using the Mahalanobis distance formula can take into account the correlation between variables and improve the accuracy of the calculation results.

[0193] In some examples, the third distance parameter is obtained by performing calculation based on the second distance parameter and the first weight, and the processing module 20 is further configured to:

[0194] After subtracting the state matrix of the visual obstacle from the state matrix of the second candidate obstacle, the obtained matrix is ​​multiplied by the first weight and then transposed to obtain a second transposed matrix of the visual obstacle and the second candidate obstacle;

[0195] Obtain an inverse matrix of a second covariance matrix between the visual obstacle and the second candidate obstacle according to the second transposed matrix;

[0196] The third distance parameter is obtained by multiplying the second transposed matrix by the inverse matrix of the second covariance matrix.

[0197] Combining the above example, the formula for calculating the third distance parameter is as follows:

[0198]

[0199] The adjusted weight is In the case of , the calculation result of the third distance parameter is .

[0200] By improving the Mahalanobis distance formula, adding weights to the original Mahalanobis distance formula, and adjusting the weights according to different scenarios, the Mahalanobis distance calculation results can be made more in line with the actual operation needs of the vehicle.

[0201] In some examples, the first distance parameter is obtained by calculating the third distance parameter and the number of times the second candidate obstacle is associated using a nonlinear activation function. The processing module 20 is further configured to:

[0202] The number of times the second candidate obstacle is associated is subtracted and divided by a preset parameter. The result is calculated using a nonlinear activation function to obtain an optimized value.

[0203] The first distance parameter is obtained by multiplying the result obtained by subtracting the preset value from the optimized value by the third distance parameter.

[0204] Specifically, considering that the longer a track is associated with a fixed obstacle, the greater the probability that they are the same obstacle, it is necessary to "encourage" them to continue to be associated. Therefore, the formula for calculating the first distance parameter is as follows:

[0205]

[0206] in, represents a nonlinear activation function, Indicates the number of times the second candidate obstacle is associated, Indicates the preset parameters. represents the third distance parameter calculated in the previous example. Represents the first distance parameter. Specifically, the nonlinear activation function is: ; In the function In combination with the previous example, for example, the second candidate obstacle is associated 4 times, and the preset parameter is 1.6, then 2.5, substituting it into the nonlinear activation function, we can get ; .

[0207] In practice, the optimization value is calculated using a nonlinear activation function, which is calculated by subtracting the number of times the second candidate obstacle has been associated and dividing it by a preset parameter. The result is then subtracted from the preset value and multiplied by the third distance parameter to obtain the first distance parameter. As the association binding time increases, the value of the distance parameter decreases. At the same time, the inherent characteristics of the Sigmoid function ensure that when bind_cnt is small, the "encouragement" of association behavior is not obvious. As bind_cnt gradually increases, the compensation increases. Even when bind_cnt is infinite, the Mahalanobis distance score will not be negative.

[0208] In some examples, the processing module 20 is further configured to:

[0209] Determine whether the third distance parameter corresponding to the second candidate obstacle corresponding to the minimum distance parameter is a minimum value;

[0210] If the corresponding third distance parameter is not the minimum value, the number of times the second candidate obstacle corresponding to the minimum distance parameter is associated is updated according to the exponential function, the first distance parameter is recalculated and screened, and the second candidate obstacle corresponding to the minimum distance parameter obtained through screening is used as the target obstacle;

[0211] If the corresponding third distance parameter is the minimum value, the second candidate obstacle corresponding to the minimum distance parameter is used as the target obstacle.

[0212] To avoid the situation where the track is associated with the current target obstacle while the target obstacle is not present and compensation is continuously being applied to the current target obstacle, even if the target obstacle subsequently appears, the calculated first distance parameter will be greater than the compensated first distance parameter of the current target obstacle, making association impossible. Therefore, before determining the target obstacle, the calculated third distance parameter of each candidate obstacle before compensation is constantly compared.

[0213] In practical applications, combined with the previous example, the smallest distance parameter among the first distance parameters calculated is , and the corresponding third distance parameter of the second candidate obstacle is , the third distance parameter Compare the third distance parameter with the remaining third distance parameters. If the third distance parameter is not the minimum value, and the minimum value of the third distance parameter is 16.35, then use the exponential function to update the number of associations for the second candidate obstacle with a third distance parameter of 16.75. Recalculate the first distance parameter using the distance parameter formula and perform screening. The second candidate obstacle corresponding to the minimum distance parameter obtained in this screening is used as the target obstacle. If the third distance parameter is the minimum value, then the second candidate obstacle corresponding to the minimum distance parameter is used as the target obstacle.

[0214] In some examples, the number of times the second candidate obstacle is associated corresponding to the minimum distance parameter is updated according to the exponential function, and the processing module 20 is further configured to:

[0215] The number of association errors of the second candidate obstacle corresponding to the minimum distance parameter is increased by one and then the exponent is taken. The result is multiplied by the number of times the second candidate obstacle is associated and rounded down to obtain the new number of times the second candidate obstacle is associated.

[0216] Specifically, the formula for updating the number of times the second candidate obstacle is associated corresponding to the minimum distance parameter is: ,in, Indicates the number of association errors; for example, combined with the above example, assuming that the current number of association errors is 1, add 1 to the current number of association errors to get , substituting into the formula, the new number of associations of the second candidate obstacle is By updating the number of times the second candidate obstacle is associated, the value of bind_cnt can be quickly reduced, which is conducive to switching from the wrong associated target to the correct associated target in time.

[0217] In some examples, after the association probability of the radar of the first candidate obstacle is zero, the processing module 20 is further configured to:

[0218] The first obstacle candidate with an association probability of zero is used as the third obstacle candidate, and the third obstacle candidate is returned to execute the above steps starting from the screening process of the plurality of radar obstacles.

[0219] Specifically, if the association probability for the radar of the first candidate obstacle is zero, it indicates that the first candidate obstacle does not meet the current requirements. It is then selected as the third candidate obstacle and the initial steps of the multiple radar obstacle screening process are repeated. When the distance is far, the visual obstacle attributes have not converged and fluctuate significantly. First, the target obstacle radar with the closest attributes, radar three, is selected for association to ensure the earliest possible track appearance, which facilitates vehicle control. If, during the screening process, radar three's attributes are determined to be non-compliant, the association probability is zero, and radar one, the target obstacle, is selected from the remaining candidate obstacles for association. Re-including the first candidate obstacle with a zero association probability as the third candidate obstacle in the screening process prevents any association errors caused by data anomalies in the early stages, allowing them to be corrected in the next round of screening. This makes the data processing process more robust, reduces the impact of erroneous data on the final results, and enhances the system's ability to cope with complex situations.

[0220] The present embodiment provides a vehicle track updating device that can execute the method provided in the above method embodiment. Its implementation principle and technical effects are similar and will not be described in detail in this embodiment.

[0221] Figure 7 This application provides a structural diagram of an electronic device. Figure 7 As shown, the electronic device 70 provided in this embodiment includes: at least one processor 701 and a memory 702. Optionally, the device 70 further includes a communication component 703. The processor 701, the memory 702 and the communication component 703 are connected via a bus 704.

[0222] During the specific implementation process, at least one processor 701 executes the computer-executable instructions stored in the memory 702, so that the at least one processor 701 performs the above method.

[0223] The specific implementation process of the processor 701 can be found in the above method embodiment. Its implementation principle and technical effects are similar and will not be repeated here in this embodiment.

[0224] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the present invention may be directly executed by a hardware processor or by a combination of hardware and software modules within the processor.

[0225] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage.

[0226] A bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be categorized as address buses, data buses, and control buses. For ease of illustration, the buses in the drawings of this application are not limited to just one bus or just one type of bus.

[0227] The present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0228] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0229] The readable storage medium may be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium may be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0230] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be an integral part of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0231] The division of units is merely a logical functional division; actual implementations may employ alternative divisions, such as combining or integrating multiple units or components into another system, or omitting or disabling certain features. Furthermore, any direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units, either through an interface, electrical, mechanical, or other means.

[0232] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0233] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0234] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0235] Those skilled in the art will appreciate that all or part of the steps in the above-described method embodiments can be implemented using hardware associated with program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0236] Finally, it should be noted that those skilled in the art will readily identify other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art not disclosed herein. The present invention is not limited to the precise structure described above and illustrated in the accompanying drawings, and various modifications and variations may be made without departing from the scope thereof. The scope of the present invention is limited solely by the appended claims.

Claims

1. A vehicle track updating method, characterized in that: include: Obtaining a visual obstacle and multiple radar obstacles corresponding to the vehicle, and screening the multiple radar obstacles to obtain multiple first candidate obstacles; determining an association probability between each first obstacle candidate and the visual obstacle based on the attribute information of the plurality of first obstacle candidate objects and the attribute information of the visual obstacle, and determining a plurality of second obstacle candidate objects based on the association probability; determining a first distance parameter of each second candidate obstacle, and determining a target obstacle from the plurality of second candidate obstacles based on the first distance parameter; The track of the vehicle is updated based on the target obstacle.

2. The method according to claim 1, characterized in that The updating of the vehicle's track based on the target obstacle includes: Determining whether the target obstacle is a currently associated target obstacle; If the target obstacle is not the currently associated target obstacle, the target obstacle is associated with the vehicle's track to obtain a new track, the association relationship between the new track and the visual obstacle is updated; the attributes of the new track are filtered and smoothed; and the vehicle's track is updated to the new track; If the target obstacle is the currently associated target obstacle, the target obstacle is associated with the vehicle's track to obtain a new track, the association relationship between the new track and the visual obstacle is updated; and the vehicle's track is updated to the new track.

3. The method according to claim 1, characterized in that The screening process of the plurality of radar obstacles to obtain a plurality of first candidate obstacles includes: determining, for any one of the multiple radar obstacles, a longitudinal distance error and a lateral distance error between the radar obstacle and the visual obstacle, and determining a distance error value corresponding to the radar obstacle based on the longitudinal distance error and the lateral distance error; Determining whether the distance error value is less than a preset error value; If the distance error value is less than a preset error value, the radar obstacle is taken as a first candidate obstacle; If the distance error value is not less than the preset error value, the radar obstacle is removed.

4. The method according to claim 3, characterized in that The determining, based on the longitudinal distance error and the lateral distance error, a distance error value corresponding to the radar obstacle includes: The longitudinal distance error is divided by the longitudinal preset parameter and then squared to obtain a longitudinal distance error value; the lateral distance error is divided by the lateral preset parameter and then squared to obtain a lateral distance error value; The longitudinal distance error value and the lateral distance error value are summed and calculated to obtain a distance error value corresponding to the radar obstacle.

5. The method according to claim 1, wherein The determining, based on the attribute information of the first candidate obstacle and the attribute information of the visual obstacle, an association probability between each first candidate obstacle and the visual obstacle, and determining a plurality of second candidate obstacles based on the association probability, includes: For any one of the plurality of first candidate obstacles, determining whether the number of attribute information of the first candidate obstacle and the attribute information of the visual obstacle that are consistent reaches a preset number; If the attribute consistency reaches a preset number of items, the association probability of the first candidate obstacle is not zero, and the first candidate obstacle with a non-zero association probability is used as the second candidate obstacle; If the attribute consistency does not reach the preset number of items, the association probability of the first candidate obstacle is zero.

6. The method according to claim 5, characterized in that The determining of the first distance parameter of each second candidate obstacle includes: For any one of the plurality of second candidate obstacles and the visual obstacle, determining a state quantity matrix of the visual obstacle and a state quantity matrix of the second candidate obstacle, and determining a second distance parameter of the second candidate obstacle based on the state quantity matrix of the visual obstacle and the state quantity matrix of the second candidate obstacle; wherein the state quantities represent position data and speed data of the second candidate obstacle and the visual obstacle; Adjusting the preset weights of different directions based on historical experience to obtain a first weight; wherein the different directions include longitudinal position, transverse position, longitudinal speed, and transverse speed; Calculating according to the second distance parameter and the first weight to obtain a third distance parameter; Obtaining the number of times the second candidate obstacle is associated, and calculating the first distance parameter using a nonlinear activation function based on the third distance parameter and the number of times the second candidate obstacle is associated; The minimum distance parameter is selected from the first distance parameters.

7. The method according to claim 6, characterized in that The determining, based on the state matrix of the visual obstacle and the state matrix of the second candidate obstacle, a second distance parameter of the second candidate obstacle includes: Subtracting the state matrix of the visual obstacle from the state matrix of the second candidate obstacle, and transposing the resulting matrix to obtain a first transposed matrix of the visual obstacle and the second candidate obstacle; Obtaining an inverse matrix of a first covariance matrix between the visual obstacle and the second candidate obstacle according to the first transposed matrix; The first transposed matrix is ​​multiplied by the inverse matrix of the first covariance matrix to obtain a second distance parameter.

8. The method according to claim 7, characterized in that The calculating according to the second distance parameter and the first weight to obtain a third distance parameter includes: Subtracting the state matrix of the visual obstacle from the state matrix of the second candidate obstacle, multiplying the resulting matrix by the first weight and then transposing the matrix to obtain a second transposed matrix of the visual obstacle and the second candidate obstacle; Obtaining an inverse matrix of a second covariance matrix between the visual obstacle and the second candidate obstacle according to the second transposed matrix; The second transposed matrix is ​​multiplied by the inverse matrix of the second covariance matrix to obtain a third distance parameter.

9. The method according to claim 8, characterized in that The step of calculating the first distance parameter based on the third distance parameter and the number of times the second candidate obstacle is associated with the first candidate obstacle by using a nonlinear activation function includes: The result obtained by subtracting the number of times the second candidate obstacle is associated and dividing it by a preset parameter is calculated using a nonlinear activation function to obtain an optimized value; The first distance parameter is obtained by multiplying the result obtained by subtracting the optimized value from the preset value by the third distance parameter.

10. The method according to claim 6, characterized in that The method further comprises: Determining whether the third distance parameter corresponding to the second candidate obstacle corresponding to the minimum distance parameter is a minimum value; If the corresponding third distance parameter is not the minimum value, then after updating the number of times the second candidate obstacle corresponding to the minimum distance parameter is associated according to the exponential function, the first distance parameter is recalculated and screened, and the second candidate obstacle corresponding to the minimum distance parameter obtained by screening is used as the target obstacle; If the corresponding third distance parameter is a minimum value, the second candidate obstacle corresponding to the minimum distance parameter is used as the target obstacle.

11. The method according to claim 10, characterized in that The updating, according to the exponential function, of the number of times the second candidate obstacle is associated corresponding to the minimum distance parameter includes: The number of association errors of the second candidate obstacle corresponding to the minimum distance parameter is increased by one, and the resulting value is multiplied by the number of times the second candidate obstacle is associated, and the result is rounded down to obtain a new number of times the second candidate obstacle is associated.

12. The method according to claim 5, characterized in that After the association probability of the radar of the first candidate obstacle is zero, the method further includes: The first obstacle candidate with an association probability of zero is used as a third obstacle candidate, and the third obstacle candidate is returned to execute the above steps starting from screening the plurality of radar obstacles.

13. A vehicle track updating device, characterized in that: include: an acquisition module, configured to acquire a visual obstacle and a plurality of radar obstacles corresponding to the vehicle, and screen the plurality of radar obstacles to obtain a plurality of first candidate obstacles; a processing module, configured to determine an association probability between each first candidate obstacle and the visual obstacle based on the attribute information of the plurality of first candidate obstacles and the attribute information of the visual obstacle, and determine a plurality of second candidate obstacles based on the association probability; The processing module is further configured to determine a first distance parameter of each second candidate obstacle, and determine a target obstacle from the plurality of second candidate obstacles based on the first distance parameter; The processing module is further configured to update the track of the vehicle based on the target obstacle.

14. An electronic device, characterized in that: include: Memory, processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor performs the method according to any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 12 when executed by a processor.

16. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 12 when executed by a processor.

Images