Pedestrian trajectory prediction model testing method and system based on lidar

Abstract

The present invention discloses a laser radar-based pedestrian trajectory prediction model testing method and system, wherein the method includes the following steps: using a laser radar to detect 3D targets in a selected real environment and filter pedestrian data; matching the perceived pedestrians using a multi-target tracking algorithm to construct a two-dimensional target tracking array; performing a frame extraction operation on the two-dimensional pedestrian target tracking array according to the input frame spacing of the model to be tested to obtain a sampling array; performing data completion on the sampling array to obtain a test array; sorting the test array corresponding to each pedestrian in a certain order to obtain a pedestrian trajectory data file; inputting the pedestrian trajectory data file into the pedestrian trajectory prediction model to be tested; comparing the predicted results with the pedestrian trajectory true value, and evaluating the accuracy and real-time performance of the model. This invention can improve the reliability of the performance testing of pedestrian trajectory prediction models before application in real environments.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a laser radar-based pedestrian trajectory prediction model testing method and system. Background Art

[0002] With the continuous development of autonomous driving technology and intelligent monitoring systems, the demand for pedestrian trajectory prediction models is increasing. These models need to accurately predict pedestrian movement trajectories in various environments to ensure that the system accurately understands pedestrian behavior and can make smarter and safer decisions. However, due to the limited size of open pedestrian trajectory datasets, many datasets may only cover limited environments or scenarios, such as specific public places or parts of cities. Open pedestrian trajectory prediction models cannot be verified for specific application scenarios, which limits the model's generalization ability. At the same time, the trajectory data in the dataset may contain errors or inconsistencies, especially when collecting data with cameras. The data errors are large due to low camera accuracy and improper data processing or trajectory extraction technology. Summary of the Invention

[0003] In response to the defect that the existing developed pedestrian trajectory prediction models cannot perform performance verification for specific application scenarios, the present invention provides a lidar-based pedestrian trajectory prediction model testing method and system that can realize the collection and processing of customized scene data and improve the prediction accuracy and inference speed of different models.

[0004] The technical solution adopted in the present invention is:

[0005] A method for testing a pedestrian trajectory prediction model based on laser radar is provided, comprising the following steps:

[0006] S1. Continuously perform 3D object detection in a selected real-world environment using LiDAR over a period of time and at a certain frequency, and filter out pedestrian data, including timestamps and pedestrian coordinates.

[0007] S2. Match the perceived pedestrian detection results using a multi-target tracking algorithm, assign pedestrian identification tags, and combine the same pedestrian data in ascending timestamp order based on the pedestrian identification tags to obtain a two-dimensional target tracking array for the pedestrian;

[0008] S3, performing a frame sampling operation on the pedestrian target tracking two-dimensional array according to the input frame spacing of the model to be tested to obtain a sampling array;

[0009] S4, completing the data of the sampling array to obtain a test array;

[0010] S5. Sort the test array corresponding to each pedestrian in a certain order to obtain a pedestrian trajectory data file;

[0011] S6. Input the pedestrian trajectory data file into the pedestrian trajectory prediction model to be tested according to the observation time and frame number of the model to be tested, and save the prediction result to the prediction result database;

[0012] S7. Compare the predicted result database with the true value of pedestrian trajectory and evaluate the accuracy and real-time performance of the model.

[0013] Following the above technical solution, in step S1, the lidar detects the 3D targets in the selected real environment through the CenterPoint network, and eliminates the identified vehicles based on the size of the identified objects, retaining only pedestrian data.

[0014] Following the above technical solution, in step S4, starting from the initial timestamp, the sampling array is checked backward in units of frame spacing to see if there are any missing frames. If there are any missing frames, linear interpolation is performed.

[0015] Following the above technical solution, in step S5, the test array of each pedestrian is specifically based on the timestamp as the first sorting element and the pedestrian identity as the second sorting element, and the pedestrian trajectory data is sorted in ascending order; finally, the pedestrian trajectory data is stored as a txt file in the format of [timestamp, pedestrian identity id, pedestrian horizontal coordinate x, pedestrian vertical coordinate y].

[0016] Following the above technical solution, in step S7, the predicted trajectory in the prediction result database is compared with the corresponding part of the pedestrian trajectory true value at the same time, and the average distance error ADE and the final distance error FDE are calculated.

[0017] Following the above technical solution, in step S8, the time before the pedestrian historical trajectory data in the pedestrian trajectory data file is input into the model to be tested is set as the start time, and the time when the predicted trajectory is output from the model to be tested is set as the end time, and multiple groups of time pairs corresponding to multiple moments are obtained. The real-time performance is judged by the average of the time differences of the multiple groups of time pairs.

[0018] Following the above technical solution, in step S1, 3D object detection is specifically used to output the pedestrian / vehicle 3D bounding box and center point parameters (x, y, z, l, w, h), where (x, y, z) are the coordinates of the center point of the object, and l, w, and h are the length, width, and height of the object's 3D bounding box, respectively. The pedestrian / vehicle size is calculated by l×w×h. When the size is greater than a certain threshold, it is defined as a vehicle and the detected vehicle results are discarded; otherwise, it is a pedestrian.

[0019] The present invention also provides a pedestrian trajectory prediction model testing system based on laser radar, comprising:

[0020] The pedestrian data acquisition module is used to continuously detect 3D objects in the selected real environment using LiDAR over a period of time and at a certain frequency, and filter out pedestrian data, including timestamps and pedestrian coordinates;

[0021] The pedestrian data processing module is used to match the perceived pedestrian detection results through a multi-target tracking algorithm, assign pedestrian identification tags, and combine the same pedestrian data in ascending timestamp order based on the pedestrian identification tags to obtain the pedestrian's target tracking two-dimensional array; perform frame extraction on the pedestrian target tracking two-dimensional array based on the input frame spacing of the model to be tested to obtain a sampling array; perform data completion on the sampling array to obtain a test array; and sort the test array corresponding to each pedestrian in a certain order to obtain a pedestrian trajectory data file;

[0022] The model prediction module is used to input the pedestrian trajectory data file into the pedestrian trajectory prediction model to be tested based on the observation time and frame number of the model to be tested, and save the prediction results to the prediction result database; compare the prediction result database with the pedestrian trajectory true value, and evaluate the accuracy and real-time performance of the model.

[0023] Following the above technical solution, the pedestrian data processing module is specifically based on the test array of each pedestrian, with the timestamp as the first sorting element and the pedestrian identity as the second sorting element, sorted in ascending order; finally, the pedestrian trajectory data is stored as a txt document in the format of [timestamp, pedestrian identity id, pedestrian horizontal coordinate x, pedestrian vertical coordinate y].

[0024] The present invention also provides a computer storage medium storing a computer program executable by a processor, wherein the computer program executes the laser radar-based pedestrian trajectory prediction model testing method described in the above technical solution.

[0025] The beneficial effects of the present invention are as follows: the present invention obtains and processes pedestrian data in a real environment through a lidar, and combines it into an array according to certain rules as test data. The pedestrian trajectory prediction model to be tested is tested by using the test data, which can better evaluate the accuracy and real-time performance of the model.

[0026] Furthermore, by detecting 3D targets in the selected real environment through the CenterPoint network, vehicles and pedestrians can be quickly identified based on the size of the targets.

[0027] Furthermore, in order to ensure the quality and real-time nature of the collected data, high-precision lidar can be used to collect data.

[0028] Furthermore, starting from the initial timestamp, the sampling array is checked backward in units of frame spacing to see if there are any missing frames. If there are any missing frames, linear interpolation is performed to compensate for the impact of sensor missed detection.

[0029] Furthermore, the pedestrian trajectory data file is input into the pedestrian trajectory prediction model to be tested according to the observation duration and frame number of the model to be tested. This method can be applied to all pedestrian trajectory prediction models, especially when the input frames of each prediction model have different observation durations.

[0030] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0032] Figure 1A This is a flowchart of a method for testing a pedestrian trajectory prediction model based on laser radar according to an embodiment of the present invention;

[0033] Figure 1B is a flowchart of a method for testing a pedestrian trajectory prediction model based on laser radar according to another embodiment of the present invention;

[0034] Figure 2 is a flow chart of performing linear interpolation on data according to an embodiment of the present invention;

[0035] Figure 3 This is a flow chart for determining the accuracy of a model to be tested according to an embodiment of the present invention;

[0036] Figure 4 This is a flow chart of determining the real-time performance of a model to be tested according to an embodiment of the present invention;

[0037] Figure 5 This is a specific judgment flow chart of a pedestrian trajectory prediction model testing method based on lidar in one embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0039] It should be noted that the illustrations provided in the embodiments of the present invention are only schematic illustrations of the basic concept of the present invention. Therefore, the drawings only show components related to the present invention and are not drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component can be changed at will, and the component layout type may also be more complicated.

[0040] In the present invention, it should also be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer" and the like are used to indicate positions or locations based on those shown in the accompanying drawings. These terms are intended solely to facilitate the description of the present application and to simplify the description. They are not intended to indicate or imply that the devices or components referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present application. Furthermore, the terms "first" and "second" are used solely for descriptive and distinguishing purposes and should not be construed as indicating or implying relative importance.

[0041] like Figure 1A As shown, the pedestrian trajectory prediction model testing method based on laser radar in an embodiment of the present invention includes the following steps:

[0042] S101, continuously performing 3D target detection in a selected real environment over a period of time and at a certain frequency using a laser radar, and filtering out pedestrian data, including timestamps and pedestrian coordinates;

[0043] S102: Match the perceived pedestrian detection results using a multi-target tracking algorithm, assign a pedestrian identity, and combine the same pedestrian data in ascending order of timestamps based on the pedestrian identity to obtain a two-dimensional target tracking array for the pedestrian;

[0044] S103, performing a frame sampling operation on the pedestrian target tracking two-dimensional array according to the input frame spacing of the model to be tested to obtain a sampling array;

[0045] S104, completing the data of the sampling array to obtain a test array;

[0046] S105, sorting the test array corresponding to each pedestrian in a certain order to obtain a pedestrian trajectory data file;

[0047] S106, inputting the pedestrian trajectory data file into the pedestrian trajectory prediction model to be tested according to the observation time and frame number of the model to be tested, and saving the prediction result to the prediction result database;

[0048] S107: Compare the prediction result database with the pedestrian trajectory true value, and evaluate the accuracy and real-time performance of the model.

[0049] Although the pedestrian trajectory prediction model has excellent test performance under public datasets, it is difficult to ensure the performance of the pedestrian trajectory prediction model in the selected real environment because the public datasets only cover limited scenarios and the data quality errors of the datasets are large. In this embodiment, a pedestrian trajectory prediction model testing method is designed. Through high-precision and real-time target detection, target tracking, and trajectory prediction processes, the accuracy and real-time performance of the pedestrian trajectory prediction model under self-selected real environment data are tested, which supplements and improves the test data source and testing method of the pedestrian trajectory prediction model.

[0050] Because the public dataset covers a single and limited scene, the present invention tests the pedestrian trajectory prediction model based on a self-selected real environment; at the same time, in order to ensure the quality and real-time nature of the collected data, a high-precision 128-line lidar can be used for data collection.

[0051] like Figure 1B and Figure 5 As shown, another embodiment of the present invention is a pedestrian trajectory prediction model testing method based on laser radar, which mainly includes the following steps:

[0052] S1. Continuously perform 3D object detection using the CenterPoint algorithm, which features high detection accuracy and fast inference speed, over a period of time. Vehicles are eliminated based on the size of the objects detected, and pedestrian data (timestamp, pedestrian coordinate x, pedestrian coordinate y) detected over a period of time is recorded.

[0053] S2. Match the perceived pedestrian detection results through the AB3DMOT multi-target tracking algorithm and assign a pedestrian ID. According to the pedestrian ID, multiple pairs of (timestamp, pedestrian ID, pedestrian coordinate x, pedestrian coordinate y) of the same pedestrian at different times are combined in ascending order of timestamp to obtain the target tracking result bank of the pedestrian ID. i [] Two-dimensional array.

[0054] S3, in order to ensure that the method is applicable to all pedestrian trajectory prediction models with different input frame time intervals, traverse i, according to the input frame time interval of the model to be tested, for each bank i []Array performs frame extraction operation to obtain the corresponding sampled_bank i [] array.

[0055] S4, in order to compensate for the impact of sensor missed detection, traverse i, for each sampled_bank i []Array to complete the data to get traj i [] array.

[0056] S5, the traj corresponding to each pedestrian i[] Sort the pedestrian trajectory data files in a certain order.

[0057] S6. To ensure that the method is applicable to all pedestrian trajectory prediction models with different observation durations of input frames, according to the observation duration N_obs frames of the model to be tested, the data corresponding to N_obs frames in the pedestrian trajectory data file are input into the pedestrian trajectory prediction model to be tested for inference, and the prediction results are saved in the prediction result database.

[0058] S7. Compare the predicted result database with the true value of the pedestrian trajectory, and calculate ADE and FDE to evaluate the accuracy of the model.

[0059] S8. Record the initial time T_begin when the pedestrian trajectory data file is input into the pedestrian trajectory prediction model to be tested and the time T_end when the prediction result is output to evaluate the real-time performance of the model.

[0060] Furthermore, step S1 can specifically perform 3D target detection at a frequency of 10Hz. In order to obtain a relatively large data scale to fully verify the performance of the model, target detection is performed within a time range of more than one hour; the pedestrian / vehicle 3D bounding box and center point parameters (x, y, z, l, w, h) are output through 3D target detection, where (x, y, z) are the coordinates of the center point of the object, and l, w, h are the length, width, and height of the object's 3D bounding box respectively; the pedestrian / vehicle size is calculated as l×w×h. When the size is greater than a certain threshold, it is defined as a vehicle and the detected vehicle result is discarded; otherwise, it is a pedestrian. Since the predicted pedestrian motion trajectory is a bird's-eye view, the coordinate z is ignored, and the pedestrian detection result is retained as (timestamp timestamp, pedestrian coordinate x, pedestrian coordinate y) pair.

[0061] Furthermore, step S3 specifically extracts frames according to the input frame time interval T0 s (such as T0 = 0.4) of the model to be tested. i [] Extract frames at intervals of T0, that is, extract all the data corresponding to the initial timestamp and the timestamp whose difference from the initial timestamp is a multiple of T0, and obtain the bank after frame extraction i [] array sampled_bank i [] Two-dimensional array.

[0062] Furthermore, step S4 specifically first checks sampled_bank i [] Check whether there are any missing frames in the array, starting from the initial timestamp, and checking every T0s backward to avoid the impact of missed detection by the lidar; if there are any missing frames, linear interpolation operation is performed on the pedestrian coordinates x and y. The linear interpolation operation can supplement the missing frame data with relatively accurate accuracy, and finally obtain the traj after data completion i [] Two-dimensional array.

[0063] Furthermore, step S5 is specifically based on the traj of each pedestrian id i [] array, with timestamp as the first sorting element and pedestrian id as the second sorting element, sorted in ascending order; finally, the pedestrian trajectory data is stored as a txt document in the format of [timestamp, pedestrian id, pedestrian coordinate x, pedestrian coordinate y]; the specific txt data file format is: [initial timestamp timestamp, 1, x coordinate of id 1, y coordinate of id 1], [initial timestamp timestamp, 2, x coordinate of id 2, y coordinate of id 2], ..., [initial timestamp timestamp+T0, 1, x coordinate of id 1, y coordinate of id 1], [initial timestamp timestamp+T0, 2, x coordinate of id 2, y coordinate of id 2], ...

[0064] Furthermore, in step S7, the predicted trajectory in the prediction result database is specifically compared with the portion of the pedestrian trajectory true value corresponding to the same time, and the average distance error (ADE) and the final distance error (FDE) are calculated.

[0065] Furthermore, in step S8, the time before the pedestrian historical trajectory data in the pedestrian trajectory data file is input into the model to be tested is set as T_begin, and the time when the predicted trajectory is output from the model to be tested is set as T_end, and multiple groups (T_begin, T_end) corresponding to multiple time points are obtained. Where N is the number of samples; the real-time performance is judged by the value of T. To make the obtained real-time performance credible, N ≥ 100 is selected.

[0066] The linear interpolation process in step S4 is as follows: Figure 2 As shown, the following steps are included:

[0067] S201. Locate the locations where missing values ​​exist in the data set.

[0068] S202. For each missing value, find the nearest known data point before and after it.

[0069] S203, using the numerical value and position information of the known data points, calculate the slope k between the two points, where

[0070]

[0071] S204. Use the calculated slope to estimate the missing value through a linear relationship, where missing value = the value of known point 1 + (slope k × (x - the position of known point 1)).

[0072] S205. When there is no missing value, the linear interpolation ends. Otherwise, the above steps are repeated to perform linear interpolation on other missing values.

[0073] The process of determining the accuracy of the model to be tested in step S7 is as follows: Figure 3 As shown, the following steps are included:

[0074] S301: Compare the predicted trajectory in the prediction result database with the portion of the pedestrian trajectory true value corresponding to the same time.

[0075] S302: Calculate the average distance error (ADE). Where N represents the number of moments in the trajectory, x i and y i are the horizontal and vertical coordinates of the predicted position at the i-th moment, and x and y are the horizontal and vertical coordinates of the actual position at the i-th moment.

[0076] S303: Calculate the final distance error (FDE). Among them, x f and y f are the horizontal and vertical coordinates of the predicted final position, and x and y are the horizontal and vertical coordinates of the actual final position.

[0077] S304: Determine the accuracy of the model to be tested based on the calculated ADE and FDE.

[0078] The process of judging the real-time performance of the model to be tested in step S7 is as follows: Figure 4 As shown, the following steps are included:

[0079] S401 : Set the time before the pedestrian historical trajectory data in the pedestrian trajectory data file is input into the model to be tested as T_begin.

[0080] S402: The time when the predicted trajectory is output from the model to be tested is set as T_end.

[0081] S403. Obtain multiple groups (T_begin, T_end) corresponding to multiple moments.

[0082] S404, calculation Where N is the sample size, and N must be greater than 100.

[0083] S405: Determine the real-time performance of the model to be tested based on the calculated T.

[0084] The laser radar-based pedestrian trajectory prediction model testing system of the embodiment of the present invention is mainly used to implement the above method embodiment, and the system includes:

[0085] The pedestrian data acquisition module is used to continuously detect 3D objects in the selected real environment using LiDAR over a period of time and at a certain frequency, and filter out pedestrian data, including timestamps and pedestrian coordinates;

[0086] The pedestrian data processing module is used to match the perceived pedestrian detection results through a multi-target tracking algorithm, assign pedestrian identification tags, and combine the same pedestrian data in ascending timestamp order based on the pedestrian identification tags to obtain the pedestrian's target tracking two-dimensional array; perform frame extraction on the pedestrian target tracking two-dimensional array based on the input frame spacing of the model to be tested to obtain a sampling array; perform data completion on the sampling array to obtain a test array; and sort the test array corresponding to each pedestrian in a certain order to obtain a pedestrian trajectory data file;

[0087] The model prediction module is used to input the pedestrian trajectory data file into the pedestrian trajectory prediction model to be tested based on the observation time and frame number of the model to be tested, and save the prediction results to the prediction result database; compare the prediction result database with the pedestrian trajectory true value, and evaluate the accuracy and real-time performance of the model.

[0088] Furthermore, the pedestrian data processing module is specifically based on the test array of each pedestrian, with the timestamp as the first sorting element and the pedestrian identity as the second sorting element, sorted in ascending order; finally, the pedestrian trajectory data is stored as a txt document in the format of [timestamp, pedestrian identity id, pedestrian horizontal coordinate x, pedestrian vertical coordinate y].

[0089] Each module is mainly used to implement each step of the above method embodiment, which will not be described in detail here.

[0090] The present application also provides a computer-readable storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a disk, an optical disk, a server, an App store, etc., on which a computer program is stored, and when the program is executed by a processor, a corresponding function is implemented. When the computer-readable storage medium of this embodiment is executed by a processor, the pedestrian trajectory prediction model testing method based on a laser radar of the method embodiment is implemented.

[0091] In summary, the present invention addresses the problem that existing pedestrian trajectory prediction models are mainly tested based on open datasets, which makes the test scenarios limited and single. It can effectively increase the number of test scenarios for pedestrian trajectory prediction models, thereby improving the reliability of performance testing of pedestrian trajectory prediction models before application in real environments.

[0092] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, or two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0093] The size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0094] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A pedestrian trajectory prediction model testing method based on laser radar, characterized in that: The following steps are involved: S1. Continuously perform 3D object detection in a selected real-world environment using LiDAR over a period of time and at a certain frequency, and filter out pedestrian data, including timestamps and pedestrian coordinates. S2. Match the perceived pedestrian detection results using a multi-target tracking algorithm, assign pedestrian identification tags, and combine the same pedestrian data in ascending timestamp order based on the pedestrian identification tags to obtain a two-dimensional target tracking array for the pedestrian; S3, performing a frame sampling operation on the pedestrian target tracking two-dimensional array according to the input frame spacing of the model to be tested to obtain a sampling array; S4, completing the data of the sampling array to obtain a test array; S5. Sort the test array corresponding to each pedestrian in a certain order to obtain a pedestrian trajectory data file; S6. Input the pedestrian trajectory data file into the pedestrian trajectory prediction model to be tested according to the observation time and frame number of the model to be tested, and save the prediction result to the prediction result database; S7. Compare the predicted trajectory in the prediction result database with the corresponding part of the pedestrian trajectory true value at the same time, calculate the average distance error ADE and the final distance error FDE, and determine the accuracy and real-time performance of the model to be tested based on the calculated ADE and FDE.

2. The laser radar-based pedestrian trajectory prediction model testing method according to claim 1, characterized in that: In step S1, the lidar detects 3D targets in the selected real environment through the CenterPoint network, and removes the identified vehicles based on the size of the identified objects, retaining only pedestrian data.

3. The laser radar-based pedestrian trajectory prediction model testing method according to claim 1, characterized in that: Specifically, step S4 starts from the initial timestamp and checks backwards in the sampling array with the frame spacing as the unit to see if there is any missing frame. If there is any missing frame, linear interpolation is performed.

4. The laser radar-based pedestrian trajectory prediction model testing method according to claim 1, characterized in that: In step S5, the test array of each pedestrian is specifically based on the timestamp as the first sorting element and the pedestrian identity as the second sorting element, and the pedestrian trajectory data is finally stored as a txt file in the format of [timestamp, pedestrian identity id, pedestrian horizontal coordinate x, pedestrian vertical coordinate y].

5. The laser radar-based pedestrian trajectory prediction model testing method according to claim 1, characterized in that: In step S8, the time before the pedestrian historical trajectory data in the pedestrian trajectory data file is input into the model to be tested is set as the start time, and the time when the predicted trajectory is output from the model to be tested is set as the end time, so as to obtain multiple groups of time pairs corresponding to multiple moments. The real-time performance is judged by the average of the time differences of the multiple groups of time pairs.

6. The laser radar-based pedestrian trajectory prediction model testing method according to claim 1, characterized in that: In step S1, 3D object detection is used to output the pedestrian / vehicle 3D bounding box and center point parameters (x, y, z, l, w, h), where (x, y, z) are the coordinates of the object's center point, and l, w, and h are the length, width, and height of the object's 3D bounding box, respectively. The size of the pedestrian / vehicle is calculated as l×w×h. When the size is greater than a certain threshold, it is defined as a vehicle and the detected vehicle result is discarded; otherwise, it is a pedestrian.

7. A pedestrian trajectory prediction model testing system based on laser radar, characterized in that: include: The pedestrian data acquisition module is used to continuously detect 3D objects in the selected real environment using LiDAR over a period of time and at a certain frequency, and filter out pedestrian data, including timestamps and pedestrian coordinates; The pedestrian data processing module is used to match the perceived pedestrian detection results through a multi-target tracking algorithm, assign pedestrian identification tags, and combine the same pedestrian data in ascending timestamp order based on the pedestrian identification tags to obtain a two-dimensional target tracking array for the pedestrian. The two-dimensional target tracking array of the pedestrian is subjected to a frame sampling operation based on the input frame spacing of the model to be tested to obtain a sampling array. The sampling array is completed to obtain a test array; the test array corresponding to each pedestrian is sorted in a certain order to obtain a pedestrian trajectory data file; The model prediction module is used to input the pedestrian trajectory data file into the pedestrian trajectory prediction model to be tested according to the observation time and frame number of the model to be tested, and save the prediction results to the prediction result database; The predicted trajectory in the prediction result database is compared with the corresponding part of the pedestrian trajectory true value at the same time, and the average distance error ADE and the final distance error FDE are calculated. The accuracy and real-time performance of the model to be tested are determined based on the calculated ADE and FDE.

8. The laser radar-based pedestrian trajectory prediction model testing system according to claim 7, characterized in that: The pedestrian data processing module specifically sorts the test array of each pedestrian in ascending order, using the timestamp as the first sorting element and the pedestrian identity as the second sorting element; Finally, the pedestrian trajectory data is stored as a txt file in the format of [timestamp, pedestrian identity ID, pedestrian horizontal coordinate x, pedestrian vertical coordinate y].

9. A computer storage medium, characterized in that A computer program that can be executed by a processor is stored therein, and the computer program executes the pedestrian trajectory prediction model testing method based on laser radar as described in any one of claims 1 to 6.

Images