Method, device, electronic device and medium for determining target detection evaluation index

By calculating the error area and position transformation of the center point target vehicle between consecutive frames, the problem of insufficient evaluation of vehicle detection jitter and timing change in the prior art is solved, and more accurate three-dimensional object detection evaluation and algorithm adjustment are achieved, which improves the vehicle Stability and accuracy of detection.

CN116543271BActive Publication Date: 2024-11-26BEIJING SENIOR SMART DRIVING TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310595590.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-05-24
Publication Date
2024-11-26
Estimated Expiration
2043-05-24

AI Technical Summary

Technical Problem

The existing 3D object detection evaluation index mAP cannot effectively evaluate the jitter situation and timing changes during vehicle detection, making it difficult for vehicles to accurately track and drive.

Method used

By calculating the error area of ​​the center point of the target vehicle between multiple consecutive frames, combining the position transformation algorithm and the intersection ratio, the jitter situation during the target vehicle detection process is evaluated, and the three-dimensional target detection algorithm is adjusted according to the error area.

Benefits of technology

More comprehensively and accurately evaluate the advantages and disadvantages of the three-dimensional object detection algorithm, improve the stability and accuracy of vehicle detection, and ensure that the vehicle can accurately track and drive.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116543271B_ABST
    Figure CN116543271B_ABST
Patent Text Reader

Abstract

The present application provides a method, device, electronic device and medium for determining target detection evaluation indicators, including: based on multiple truth value frames, determining the target truth value frame and vehicle account information corresponding to the three-dimensional detection frame of the target vehicle in each frame; based on the posture transformation algorithm, the coordinate system of the three-dimensional detection frame in each frame is transformed to determine the converted three-dimensional detection frame in each frame; based on the vehicle account information and the converted three-dimensional detection frame in each frame, the detection trajectory of the target vehicle is determined; based on the center point of the converted three-dimensional detection frame in each frame, the center point of the target truth value frame, the detection trajectory of the target vehicle and the truth value trajectory in the target truth value frame, the center point error area of ​​the target vehicle is determined. By determining the center point error area of ​​the target vehicle between multiple consecutive frames, the jitter in the target vehicle detection process is effectively evaluated, so as to more comprehensively and accurately evaluate the advantages and disadvantages of the three-dimensional target detection algorithm.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of autonomous driving technology, and in particular to methods, devices, electronic devices and media for determining target detection evaluation indicators. Background Art

[0002] 3D object detection is an important part of the autonomous driving field. Specifically, 3D object detection technology aims to use data collected by various sensors such as cameras, lidar, radar, etc. to identify roads, vehicles, pedestrians, etc. in motion, so that autonomous vehicles have environmental perception capabilities, thereby improving the safety and intelligence of autonomous vehicles. For the 3D object detection problem, the widely used evaluation index in the industry is mAP (mean Average Precision): by calculating the intersection-over-union (IoU) of the detection box of the perception output result and the annotated box of the real data, the overlap between the detection box and the true value box is determined. The calculation method of the mAP evaluation index usually starts from the obstacle itself to evaluate the pros and cons of the overall object detection algorithm.

[0003] However, the mAP method has two main disadvantages: 1) It only focuses on the evaluation of the accuracy of independent detection frames within a single timestamp, lacks consideration of the temporal changes of the vehicle, and cannot know the detection stability of the motion instance over a period of time; 2) The judgment method based on IoU is insensitive to the jitter of the vehicle. If you only focus on the stability of single-frame detection and ignore the stability of continuous frame instance detection, it will make it difficult to track the vehicle and give an accurate speed and direction, which will cause the vehicle to be unable to drive accurately. Therefore, how to effectively evaluate the jitter during vehicle detection has become a technical issue that cannot be underestimated. Summary of the invention

[0004] In view of this, the purpose of the present application is to provide a method, device, electronic device and medium for determining target detection evaluation indicators. By determining the error area of ​​the center point of the target vehicle between multiple consecutive frames, the jitter situation in the target vehicle detection process can be effectively evaluated, thereby more comprehensively and accurately evaluating the advantages and disadvantages of the three-dimensional target detection algorithm.

[0005] The present invention provides a method for determining a target detection evaluation index, and the method comprises:

[0006] Based on a plurality of pre-labeled truth value frames, determine a target truth value frame and vehicle account information corresponding to a three-dimensional detection frame of the target vehicle in each frame; wherein the target truth value frame of the three-dimensional detection frame in each frame is the same;

[0007] Based on the posture transformation algorithm, the coordinate system of the three-dimensional detection frame in each frame is transformed to determine the transformed three-dimensional detection frame in each frame;

[0008] Determine the detection track of the target vehicle based on the vehicle account information and the converted three-dimensional detection frame of each frame;

[0009] Based on the center point of the converted three-dimensional detection frame in each frame, the center point of the target true value frame, the detection trajectory of the target vehicle and the true value trajectory in the target true value frame, the center point error area of ​​the target vehicle is determined to adjust the three-dimensional target detection algorithm according to the center point error area; wherein the three-dimensional target detection algorithm is used to detect the three-dimensional detection frame of the target vehicle in each frame.

[0010] In a possible implementation manner, performing coordinate system transformation on the three-dimensional detection frame in each frame based on a posture transformation algorithm to determine the transformed three-dimensional detection frame in each frame includes:

[0011] Obtaining coordinates of eight vertices of the three-dimensional detection frame in the vehicle coordinate system;

[0012] Based on the eight vertex coordinates of the three-dimensional detection frame in the vehicle coordinate system, the rotation matrix, and the translation projection coefficients from the vehicle coordinate system to the world coordinate system, the three-dimensional detection frame after conversion in each frame is determined.

[0013] In a possible implementation manner, determining the detection track of the target vehicle based on the vehicle account information and the three-dimensional detection frame converted in each frame includes:

[0014] Based on the eight vertex coordinates of the three-dimensional detection frame after conversion in each frame in the world coordinate system, determining the center point coordinates of the three-dimensional detection frame after conversion in each frame in the world coordinate system;

[0015] Based on the vehicle account information, the center point coordinates of the converted three-dimensional detection frame in each frame in the world coordinate system are connected to determine the detection trajectory of the target vehicle.

[0016] In a possible implementation manner, determining the center point error area of ​​the target vehicle based on the center point of the three-dimensional detection frame converted in each frame, the center point of the target truth frame, the detection trajectory of the target vehicle, and the truth trajectory in the target truth frame includes:

[0017] Determine the difference between the center point of the three-dimensional detection frame converted in the first frame and the center point of the target true value frame in the first frame as the first side length;

[0018] Determine the difference between the center point of the converted three-dimensional detection frame in the second frame and the center point of the target true value frame in the second frame as the second side length;

[0019] The displacement length of the true value trajectory between the first frame and the second frame is determined as the third side length; the displacement length of the detection trajectory between the first frame and the second frame is determined as the fourth side length;

[0020] Based on the first side length, the second side length, the third side length and the fourth side length, determining a geometric area formed by the first side length, the second side length, the third side length and the fourth side length;

[0021] Based on the geometric area in every two frames, the center point error area of ​​the target vehicle is determined.

[0022] In a possible implementation manner, determining the center point error area of ​​the target vehicle based on the geometric area in every two frames includes:

[0023] The geometric areas of every two frames are summed to determine the total geometric area;

[0024] The total geometric area is averaged to determine the center point error area of ​​the target vehicle.

[0025] In a possible implementation manner, after determining the detection track of the target vehicle based on the vehicle account information and the three-dimensional detection frame converted in each frame, the determination method further includes:

[0026] Determine a heading angle difference between a heading angle of the three-dimensional detection frame in a world coordinate system corresponding to each frame and a heading angle of the target true value frame in the world coordinate system;

[0027] The heading angle errors corresponding to each frame are summed to determine the total heading angle difference;

[0028] Calculating the mean of the total heading angle differences to determine the heading angle error of the target vehicle;

[0029] The three-dimensional target detection algorithm is adjusted based on the heading angle error area of ​​the target vehicle.

[0030] In a possible implementation, the method of determining the target true value frame and vehicle account information corresponding to the three-dimensional detection frame of the target vehicle in each frame based on the pre-annotated multiple true value frames includes:

[0031] Determine an intersection-over-union ratio between each of the true value frames and the three-dimensional detection frame of the target vehicle in each frame;

[0032] The maximum intersection-and-union ratio is screened out from the multiple intersection-and-union ratios, the truth value frame corresponding to the maximum intersection-and-union ratio is determined as the target truth value frame corresponding to the three-dimensional detection frame, and the vehicle account information of the target truth value frame is determined as the vehicle account information corresponding to the three-dimensional detection frame.

[0033] The present application also provides a device for determining a target detection evaluation index, the device comprising:

[0034] A first determination module is used to determine a target truth frame and vehicle account information corresponding to a three-dimensional detection frame of the target vehicle in each frame based on a plurality of pre-labeled truth frames; wherein the target truth frame of the three-dimensional detection frame in each frame is the same;

[0035] A posture transformation module, used for performing coordinate system transformation on the three-dimensional detection frame in each frame based on a posture transformation algorithm, and determining the transformed three-dimensional detection frame in each frame;

[0036] A trajectory determination module, used to determine the detection trajectory of the target vehicle based on the vehicle account information and the three-dimensional detection frame converted in each frame;

[0037] The first error determination module is used to determine the center point error area of ​​the target vehicle based on the center point of the converted three-dimensional detection frame in each frame, the center point of the target true value frame, the detection trajectory of the target vehicle and the true value trajectory in the target true value frame, so as to adjust the three-dimensional target detection algorithm according to the center point error area; wherein the three-dimensional target detection algorithm is used to detect the three-dimensional detection frame of the target vehicle in each frame.

[0038] An embodiment of the present application also provides an electronic device, including: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the method for determining the target detection evaluation index as described above are performed.

[0039] An embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for determining the target detection evaluation index as described above are executed.

[0040] The embodiments of the present application provide a method, device, electronic device and medium for determining target detection evaluation indicators, wherein the determination method includes: determining a target truth frame and vehicle account information corresponding to a three-dimensional detection frame of a target vehicle in each frame based on a plurality of pre-labeled truth frames; wherein the target truth frame of the three-dimensional detection frame in each frame is the same; performing coordinate system transformation on the three-dimensional detection frame in each frame based on a posture transformation algorithm to determine the converted three-dimensional detection frame in each frame; determining a detection trajectory of the target vehicle based on the vehicle account information and the converted three-dimensional detection frame in each frame; determining a center point error area of ​​the target vehicle based on the center point of the converted three-dimensional detection frame in each frame, the center point of the target truth frame, the detection trajectory of the target vehicle and the truth trajectory in the target truth frame, so as to adjust the three-dimensional target detection algorithm according to the center point error area; wherein the three-dimensional target detection algorithm is used to detect a three-dimensional detection frame of the target vehicle in each frame. By determining the error area of ​​the center point of the target vehicle between multiple consecutive frames, the jitter during the target vehicle detection process can be effectively evaluated, thereby more comprehensively and accurately evaluating the pros and cons of the three-dimensional target detection algorithm.

[0041] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 A flowchart of a method for determining an evaluation index for target detection provided in an embodiment of the present application;

[0044] Figure 2 A schematic diagram of a three-dimensional detection frame and a true value frame provided in an embodiment of the present application;

[0045] Figure 3 One of the structural schematic diagrams of a device for determining target detection evaluation indicators provided in an embodiment of the present application;

[0046] Figure 4 A second structural diagram of a device for determining target detection evaluation indicators provided in an embodiment of the present application;

[0047] Figure 5A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0048] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of explanation and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn in real proportion. The flowchart used in this application shows the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowchart can be implemented out of sequence, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art, under the guidance of the content of the present application, can add one or more other operations to the flowchart, or remove one or more operations from the flowchart.

[0049] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0050] In order to enable those skilled in the art to use the contents of this application, the following implementation method is given in combination with the specific application scenario of "determining target detection evaluation indicators". For those skilled in the art, the general principles defined here can be applied to other embodiments and application scenarios without departing from the spirit and scope of this application.

[0051] First, the application scenarios to which this application is applicable are introduced. This application can be applied in the field of autonomous driving technology.

[0052] Research has found that the mAP method has two main shortcomings: 1) It only focuses on the evaluation of the accuracy of independent detection frames within a single timestamp, lacks consideration of the temporal changes of the vehicle, and cannot know the detection stability of the motion instance over a period of time; 2) The judgment method based on IoU is insensitive to vehicle jitter. If you only focus on the stability of single-frame detection and ignore the stability of continuous frame instance detection, it will make it difficult to track the vehicle, give the accurate speed and direction, and thus make the vehicle unable to drive accurately. Therefore, how to effectively evaluate the jitter during vehicle detection has become a technical issue that cannot be underestimated.

[0053] Based on this, an embodiment of the present application provides a method for determining target detection evaluation indicators. By determining the error area of ​​the center point of the target vehicle between multiple consecutive frames, the jitter situation in the target vehicle detection process can be effectively evaluated, thereby more comprehensively and accurately evaluating the advantages and disadvantages of the three-dimensional target detection algorithm.

[0054] See also Figure 1 , Figure 1 This is a flow chart of a method for determining a target detection evaluation index provided in an embodiment of the present application. Figure 1 As shown in , the determination method provided in the embodiment of the present application includes:

[0055] S101: Based on multiple pre-labeled truth frames, determine the target truth frame corresponding to the three-dimensional detection frame of the target vehicle in each frame and the vehicle account information; wherein the target truth frame of the three-dimensional detection frame in each frame is the same.

[0056] In this step, based on multiple pre-labeled true value frames, the target true value frame corresponding to the three-dimensional detection frame of the target vehicle in each frame and the vehicle account information are determined.

[0057] Among them, the three-dimensional detection frame of the target vehicle in each frame is obtained by predicting the automatic driving route of the target vehicle in each frame using a three-dimensional detection algorithm.

[0058] Here, obtain the 3D detection frame and true value frame in a continuous scene, as well as the corresponding timestamp and corresponding pose information. Usually, the true value data for 3D object detection is composed of multiple scene segments. Each scene segment stores continuous frame data for a period of time. The scene segment length is n, and the timestamp is from t=0 to t=n. The true value data contains the timestamp and pose information corresponding to each frame of the point cloud data and the annotated true value frame in each frame of the point cloud. The annotated true value frame has 8 dimensions of information, namely the coordinates of the center point of the true value frame (x, y, z), the size information of the true value frame (length, width and height), the heading angle information of the true value frame, and the instance ID of the true value frame (vehicle account information).

[0059] Among them, the 3D object detection algorithm will output the corresponding three-dimensional detection frame for each frame in the real data. The three-dimensional detection frame has seven dimensions of information, namely the coordinates of the center point of the three-dimensional detection frame (x, y, z), the size information of the three-dimensional detection frame (length, width and height), and the heading angle information of the three-dimensional detection frame.

[0060] In a possible implementation, the method of determining the target true value frame and vehicle account information corresponding to the three-dimensional detection frame of the target vehicle in each frame based on the pre-annotated multiple true value frames includes:

[0061] A: Determine the intersection-over-union ratio between each of the true value boxes and the three-dimensional detection box of the target vehicle in each frame.

[0062] Here, the IoU algorithm is used to determine the intersection over union ratio between each true value box and the 3D detection box of the target vehicle in each frame.

[0063] B: Filter out the maximum intersection-and-union ratio from the multiple intersection-and-union ratios, determine the truth value frame corresponding to the maximum intersection-and-union ratio as the target truth value frame corresponding to the three-dimensional detection frame, and determine the vehicle account information of the target truth value frame as the vehicle account information corresponding to the three-dimensional detection frame.

[0064] Here, the maximum intersection-and-union ratio is screened out from multiple intersection-and-union ratios, the truth value frame corresponding to the maximum intersection-and-union ratio is determined as the target truth value frame corresponding to the three-dimensional detection frame, and the vehicle account information of the target truth value frame is determined as the vehicle account information corresponding to the three-dimensional detection frame.

[0065] Among them, the 3D detection frame and the true value frame in the same frame are paired by IoU, and the corresponding instance ID is obtained according to the corresponding true value frame. Specifically, the IoU value of each true value frame and all detection frames is calculated in turn, and the detection frame with the largest IoU with the true value frame is selected as the target true value frame, and the vehicle account information (instance ID) of the target true value frame is assigned to the 3D detection frame.

[0066] S102: performing coordinate system transformation on the three-dimensional detection frame in each frame based on a posture transformation algorithm to determine the transformed three-dimensional detection frame in each frame.

[0067] In this step, the coordinate system of the three-dimensional detection frame in each frame is transformed according to the posture transformation algorithm to determine the transformed three-dimensional detection frame in each frame.

[0068] In a possible implementation manner, performing coordinate system transformation on the three-dimensional detection frame in each frame based on a posture transformation algorithm to determine the transformed three-dimensional detection frame in each frame includes:

[0069] a: Obtain the coordinates of the eight vertices of the three-dimensional detection frame in the vehicle coordinate system.

[0070] Here, the coordinates of the eight vertices of the three-dimensional detection box in the vehicle coordinate system are obtained.

[0071] Among them, the three-dimensional detection box is a cubic cuboid, so the eight vertex coordinates are the vertex coordinates of the cubic cuboid.

[0072] b: Based on the eight vertex coordinates of the three-dimensional detection frame in the vehicle coordinate system, the rotation matrix, and the translation projection coefficients from the vehicle coordinate system to the world coordinate system, the three-dimensional detection frame after conversion in each frame is determined.

[0073] Here, the converted three-dimensional detection frame in each frame is determined according to the eight vertex coordinates of the three-dimensional detection frame in the vehicle coordinate system, the rotation matrix, and the translation projection coefficient from the vehicle coordinate system to the world coordinate system.

[0074] Here, the coordinates of the eight vertices of the transformed three-dimensional detection frame are determined by the following formula:

[0075]

[0076] Among them, B g , B v Respectively represent the eight vertex coordinates of the 3D detection box in the world coordinate system and the vehicle coordinate system, Represents a 3x3 rotation matrix, which represents the rotational projection relationship from the vehicle body coordinate system to the world coordinate system, and T represents the translational projection coefficient from the vehicle body coordinate system to the world coordinate system.

[0077] S103: Determine the detection track of the target vehicle based on the vehicle account information and the three-dimensional detection frame converted in each frame.

[0078] In this step, the detection track of the target vehicle is determined based on the vehicle account information and the converted three-dimensional detection frame of each frame.

[0079] Here, the detection trajectory of the target vehicle is the trajectory of the target vehicle predicted by using a three-dimensional target detection algorithm.

[0080] In a possible implementation manner, determining the detection track of the target vehicle based on the vehicle account information and the three-dimensional detection frame converted in each frame includes:

[0081] (1): Based on the eight vertex coordinates of the three-dimensional detection frame after transformation in each frame in the world coordinate system, the center point coordinates of the three-dimensional detection frame after transformation in each frame in the world coordinate system are determined.

[0082] Here, according to the eight vertex coordinates of the converted three-dimensional detection frame in each frame in the world coordinate system, the center point coordinates of the converted three-dimensional detection frame in each frame in the world coordinate system are determined.

[0083] (2): Based on the vehicle account information, the center point coordinates of the converted three-dimensional detection frame in each frame in the world coordinate system are connected to determine the detection trajectory of the target vehicle.

[0084] Here, the center point coordinates of the converted three-dimensional detection frame in each frame in the world coordinate system are connected according to the vehicle account information to determine the detection trajectory of the target vehicle.

[0085] Among them, the 3D detection frames in the same continuous scene are projected to the same world coordinate system through pose transformation, and their trajectories are connected in series using instance IDs to form a complete instance trajectory of the continuous scene segment (detection trajectory of the target vehicle).

[0086] S104: Based on the center point of the three-dimensional detection frame converted in each frame, the center point of the target truth frame, the detection trajectory of the target vehicle and the truth trajectory in the target truth frame, determine the center point error area of ​​the target vehicle to adjust the three-dimensional target detection algorithm according to the center point error area.

[0087] In this step, the center point error area of ​​the target vehicle is determined based on the center point of the converted three-dimensional detection frame in each frame, the center point of the target truth frame, the detection trajectory of the target vehicle, and the truth trajectory in the target truth frame, so as to adjust the three-dimensional target detection algorithm according to the center point error area.

[0088] Among them, the true value trajectory is the pre-marked trajectory route.

[0089] Here, if the error area of ​​the center point is larger than the preset error area, the three-dimensional target detection algorithm is re-optimized and adjusted. As for how to optimize and adjust the three-dimensional target detection algorithm, any existing technical means can be used to achieve it.

[0090] The three-dimensional target detection algorithm is used to detect the three-dimensional detection frame of the target vehicle in each frame.

[0091] In a possible implementation manner, determining the center point error area of ​​the target vehicle based on the center point of the three-dimensional detection frame converted in each frame, the center point of the target truth frame, the detection trajectory of the target vehicle, and the truth trajectory in the target truth frame includes:

[0092] I: Determine the difference between the center point of the three-dimensional detection frame converted in the first frame and the center point of the target true value frame in the first frame as the first side length.

[0093] Here, the difference between the center point of the converted three-dimensional detection frame in the first frame and the center point of the target true value frame in the first frame is determined as the first side length.

[0094] II: Determine the difference between the center point of the three-dimensional detection frame converted in the second frame and the center point of the target true value frame in the second frame as the second side length.

[0095] Here, the difference between the center point of the converted three-dimensional detection frame in the second frame and the center point of the target true value frame in the second frame is determined as the second side length.

[0096] III: determining the displacement length of the true value trajectory between the first frame and the second frame as the third side length; determining the displacement length of the detection trajectory between the first frame and the second frame as the fourth side length.

[0097] Here, the displacement length of the true value trajectory between the first frame and the second frame is determined as the third side length; and the displacement length of the detection trajectory between the first frame and the second frame is determined as the fourth side length.

[0098] IV: Based on the first side length, the second side length, the third side length and the fourth side length, determine a geometric area formed by the first side length, the second side length, the third side length and the fourth side length.

[0099] Here, the geometric area is determined based on the first side length, the second side length, the third side length, and the fourth side length.

[0100] Here, if the geometry formed by the first side length, the second side length, the third side length and the fourth side length is a triangle, the geometric area is determined using the triangle area formula; if it is a quadrilateral, the geometric area is determined using the quadrilateral area formula.

[0101] V: Based on the geometric area in every two frames, the center point error area of ​​the target vehicle is determined.

[0102] Here, the center point error area of ​​the target vehicle is determined based on the geometric area under every two frames.

[0103] In a possible implementation manner, determining the center point error area of ​​the target vehicle based on the geometric area in every two frames includes:

[0104] The geometric areas of every two frames are summed to determine the total geometric area; the total geometric area is averaged to determine the center point error area of ​​the target vehicle.

[0105] Here, the geometric areas of every two frames are added to determine the total geometric area, and the total geometric area is averaged to determine the center point error area of ​​the target vehicle.

[0106] For further information, see Figure 2 , Figure 2 A schematic diagram of the three-dimensional detection frame and the true value frame provided in the embodiment of the present application. Figure 2As shown in the figure, the first frame is taken as the starting frame, the difference between the center point of the 3D detection frame and the center point of the truth frame is taken as the first side length (A), the difference between the center point of the 3D detection frame of the next frame and the center point of the truth frame is taken as the second side length (B), the moving position of the truth center point within the two time differences (the displacement length of the truth trajectory) is taken as the third side length (C), the displacement length of the detection trajectory is determined as the fourth side length (D), and the area of ​​the rectangle or triangle is calculated. This area is the geometric area of ​​the current timestamp. By analogy, the next frame is taken as the starting frame to calculate the geometric area, the geometric areas at all times are counted, and the mean is calculated as the center point error area of ​​the target vehicle.

[0107] In a possible implementation manner, after determining the detection track of the target vehicle based on the vehicle account information and the three-dimensional detection frame converted in each frame, the determination method further includes:

[0108] i: Determine the heading angle difference between the heading angle of the three-dimensional detection frame in the world coordinate system corresponding to each frame and the heading angle of the target true value frame in the world coordinate system.

[0109] In this way, the heading angle difference between the heading angle of the three-dimensional detection frame in the world coordinate system corresponding to each frame and the heading angle of the target true value frame in the world coordinate system is determined.

[0110] ii: summing up the heading angle errors corresponding to each frame to determine the total heading angle difference; averaging the total heading angle difference to determine the heading angle error of the target vehicle.

[0111] Here, the heading angle errors corresponding to each frame are summed to determine the total heading angle difference, and the total heading angle difference is averaged to determine the heading angle error of the target vehicle.

[0112] iii: Adjusting the three-dimensional target detection algorithm based on the heading angle error area of ​​the target vehicle.

[0113] Here, the three-dimensional target detection algorithm is adjusted according to the heading angle error area of ​​the target vehicle. If the heading angle error area is greater than the preset error area, the three-dimensional target detection algorithm needs to be optimized and adjusted.

[0114] In a specific embodiment, a three-dimensional detection frame and a true value frame in a continuous scene, as well as the corresponding timestamp and the corresponding pose information are obtained. Usually, the true value data for 3D object detection is composed of multiple scene segments. Each scene segment stores continuous frame data for a period of time, the scene segment length is n, and the timestamp is from t=0 to t=n. The true value data contains the timestamp and pose information corresponding to the point cloud data of each frame and the annotated true value frame in each frame of the point cloud. The annotated true value frame has 8 dimensions of information, namely the coordinates (x, y, z) of the center point of the true value frame, the size information (length, width and height) of the true value frame, the heading angle information of the true value frame, and the instance ID (vehicle account information) of the true value frame. The three-dimensional detection frame and the true value frame in the same frame are paired by IoU, and the corresponding instance ID is obtained according to the corresponding true value frame. Specifically, the IoU value of each true value frame and all detection frames is calculated in turn, and the detection frame with the largest IoU with the true value frame is selected as the target true value frame, and the vehicle account information (instance ID) of the target true value frame is assigned to the three-dimensional detection frame. According to the pose transformation algorithm, the coordinate system of the three-dimensional detection frame in each frame is transformed to determine the converted three-dimensional detection frame in each frame. The first frame is taken as the starting frame, the difference between the center point of the three-dimensional detection frame and the center point of the true value frame is taken as the first side length (A), the difference between the center point of the three-dimensional detection frame in the next frame and the center point of the true value frame is taken as the second side length (B), the moving position of the true value center point within the two time differences (the displacement length of the true value trajectory) is taken as the third side length (C), and the displacement length of the detection trajectory is determined as the fourth side length (D), and the area of ​​the rectangle or triangle is calculated. This area is the geometric area of ​​the current timestamp. By analogy, the geometric area is calculated by taking the next frame as the starting frame, and the geometric areas at all times are counted and the average is calculated as the center point error area of ​​the target vehicle, or the heading angle difference between the heading angle of the three-dimensional detection frame in the world coordinate system corresponding to each frame and the heading angle of the target true value frame in the world coordinate system is determined. The heading angle errors corresponding to each frame are summed up to determine the total heading angle difference, and the total heading angle difference is averaged to determine the heading angle error of the target vehicle. By calculating the center point trajectory difference area (center point error area) and heading angle error of the target vehicle in a certain continuous period of time between the three-dimensional detection frame and the true value frame, the midpoint detection stability and heading angle detection stability of the target vehicle are evaluated respectively, thus making up for the deficiency of the mAP evaluation index.

[0115] The present application provides a method for determining a target detection evaluation index, the method comprising: determining a target truth frame and vehicle account information corresponding to a three-dimensional detection frame of a target vehicle in each frame based on a plurality of pre-annotated truth frames; wherein the target truth frame of the three-dimensional detection frame in each frame is the same; performing a coordinate system transformation on the three-dimensional detection frame in each frame based on a posture transformation algorithm to determine the three-dimensional detection frame after the transformation in each frame; determining the detection trajectory of the target vehicle based on the vehicle account information and the three-dimensional detection frame after the transformation in each frame; determining the center point error area of ​​the target vehicle based on the center point of the three-dimensional detection frame after the transformation in each frame, the center point of the target truth frame, the detection trajectory of the target vehicle and the truth trajectory in the target truth frame, so as to adjust the three-dimensional target detection algorithm according to the center point error area; wherein the three-dimensional target detection algorithm is used to detect the three-dimensional detection frame of the target vehicle in each frame. By determining the center point error area of ​​the target vehicle between multiple consecutive frames, the jitter in the target vehicle detection process is effectively evaluated, thereby more comprehensively and accurately evaluating the pros and cons of the three-dimensional target detection algorithm.

[0116] See also Figure 3 , Figure 4 , Figure 3 One of the structural schematic diagrams of a device for determining target detection evaluation indicators provided in an embodiment of the present application; Figure 4 This is a second structural diagram of a device for determining target detection evaluation indicators provided in an embodiment of the present application. Figure 3 As shown in , the target detection evaluation index determination device 300 includes:

[0117] A first determination module 310 is used to determine a target truth frame corresponding to a three-dimensional detection frame of the target vehicle in each frame and vehicle account information based on a plurality of pre-labeled truth frames; wherein the target truth frame of the three-dimensional detection frame in each frame is the same;

[0118] A posture transformation module 320, configured to transform the coordinate system of the three-dimensional detection frame in each frame based on a posture transformation algorithm, and determine the transformed three-dimensional detection frame in each frame;

[0119] A trajectory determination module 330, configured to determine a detection trajectory of the target vehicle based on the vehicle account information and the converted three-dimensional detection frame in each frame;

[0120] The first error determination module 340 is used to determine the center point error area of ​​the target vehicle based on the center point of the converted three-dimensional detection frame in each frame, the center point of the target true value frame, the detection trajectory of the target vehicle and the true value trajectory in the target true value frame, so as to adjust the three-dimensional target detection algorithm according to the center point error area; wherein the three-dimensional target detection algorithm is used to detect the three-dimensional detection frame of the target vehicle in each frame.

[0121] Further, when the posture transformation module 320 is used to perform coordinate system transformation on the three-dimensional detection frame in each frame based on the posture transformation algorithm to determine the transformed three-dimensional detection frame in each frame, the posture transformation module 320 is specifically used to:

[0122] Obtaining coordinates of eight vertices of the three-dimensional detection frame in the vehicle coordinate system;

[0123] Based on the eight vertex coordinates of the three-dimensional detection frame in the vehicle coordinate system, the rotation matrix, and the translation projection coefficients from the vehicle coordinate system to the world coordinate system, the three-dimensional detection frame after conversion in each frame is determined.

[0124] Further, when the trajectory determination module 330 is used to determine the detection trajectory of the target vehicle based on the vehicle account information and the three-dimensional detection frame converted in each frame, the trajectory determination module 330 is specifically used to:

[0125] Based on the eight vertex coordinates of the three-dimensional detection frame after conversion in each frame in the world coordinate system, determining the center point coordinates of the three-dimensional detection frame after conversion in each frame in the world coordinate system;

[0126] Based on the vehicle account information, the center point coordinates of the converted three-dimensional detection frame in each frame in the world coordinate system are connected to determine the detection trajectory of the target vehicle.

[0127] Further, when the first error determination module 340 is used to determine the center point error area of ​​the target vehicle based on the center point of the three-dimensional detection frame converted in each frame, the center point of the target true value frame, the detection trajectory of the target vehicle, and the true value trajectory in the target true value frame, the first error determination module 340 is specifically used to:

[0128] Determine the difference between the center point of the three-dimensional detection frame converted in the first frame and the center point of the target true value frame in the first frame as the first side length;

[0129] Determine the difference between the center point of the converted three-dimensional detection frame in the second frame and the center point of the target true value frame in the second frame as the second side length;

[0130] The displacement length of the true value trajectory between the first frame and the second frame is determined as the third side length; the displacement length of the detection trajectory between the first frame and the second frame is determined as the fourth side length;

[0131] Based on the first side length, the second side length, the third side length and the fourth side length, determining a geometric area formed by the first side length, the second side length, the third side length and the fourth side length;

[0132] Based on the geometric area in every two frames, the center point error area of ​​the target vehicle is determined.

[0133] Further, when the first error determination module 340 is used to determine the center point error area of ​​the target vehicle based on the geometric area under every two frames, the first error determination module 340 is specifically used to:

[0134] The geometric areas of every two frames are summed to determine the total geometric area;

[0135] The total geometric area is averaged to determine the center point error area of ​​the target vehicle.

[0136] Further, such as Figure 4 As shown, the target detection evaluation index determination device 300 further includes a second error determination module 350, and the second error determination module 350 is further used to:

[0137] Determine a heading angle difference between a heading angle of the three-dimensional detection frame in a world coordinate system corresponding to each frame and a heading angle of the target true value frame in the world coordinate system;

[0138] The heading angle errors corresponding to each frame are summed to determine the total heading angle difference;

[0139] Calculating the mean of the total heading angle differences to determine the heading angle error of the target vehicle;

[0140] The three-dimensional target detection algorithm is adjusted based on the heading angle error area of ​​the target vehicle.

[0141] Further, when the first determination module 310 is used to determine the target true value frame and vehicle account information corresponding to the three-dimensional detection frame of the target vehicle in each frame based on the pre-labeled multiple true value frames, the first determination module 310 is specifically used to:

[0142] Determine an intersection-over-union ratio between each of the true value frames and the three-dimensional detection frame of the target vehicle in each frame;

[0143] The maximum intersection-and-union ratio is screened out from the multiple intersection-and-union ratios, the truth value frame corresponding to the maximum intersection-and-union ratio is determined as the target truth value frame corresponding to the three-dimensional detection frame, and the vehicle account information of the target truth value frame is determined as the vehicle account information corresponding to the three-dimensional detection frame.

[0144] An embodiment of the present application provides a device for determining a target detection evaluation index, the determination device comprising: a first determination module, for determining a target truth frame and vehicle account information corresponding to a three-dimensional detection frame of a target vehicle in each frame based on a plurality of pre-labeled truth frames; wherein the target truth frame of the three-dimensional detection frame in each frame is the same; a posture transformation module, for performing a coordinate system transformation on the three-dimensional detection frame in each frame based on a posture transformation algorithm, and determining the converted three-dimensional detection frame in each frame; a trajectory determination module, for determining a detection trajectory of the target vehicle based on the vehicle account information and the converted three-dimensional detection frame in each frame; a first error determination module, for determining a center point error area of ​​the target vehicle based on the center point of the converted three-dimensional detection frame in each frame, the center point of the target truth frame, the detection trajectory of the target vehicle and the truth trajectory in the target truth frame, so as to adjust the three-dimensional target detection algorithm according to the center point error area; wherein the three-dimensional target detection algorithm is used to detect the three-dimensional detection frame of the target vehicle in each frame. By determining the error area of ​​the center point of the target vehicle between multiple consecutive frames, the jitter during the target vehicle detection process can be effectively evaluated, thereby more comprehensively and accurately evaluating the pros and cons of the three-dimensional target detection algorithm.

[0145] See also Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 5 As shown in , the electronic device 500 includes a processor 510 , a memory 520 and a bus 530 .

[0146] The memory 520 stores machine-readable instructions executable by the processor 510. When the electronic device 500 is running, the processor 510 communicates with the memory 520 via the bus 530. When the machine-readable instructions are executed by the processor 510, the above-mentioned Figure 1 The steps of the method for determining the target detection evaluation index in the method embodiment shown, the specific implementation method can be found in the method embodiment, and will not be repeated here.

[0147] The present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the computer program can execute the above-mentioned Figure 1The steps of the method for determining the target detection evaluation index in the method embodiment shown, the specific implementation method can be found in the method embodiment, and will not be repeated here.

[0148] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0149] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0150] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0151] In addition, each functional unit in each embodiment of the present application 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.

[0152] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application can essentially be embodied in the form of a software product, or in other words, the part that contributes to the prior art or the part of the technical solution. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0153] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for determining target detection evaluation indicators, characterized in that: The determination method comprises: Based on multiple pre-labeled truth value frames, determine the target truth value frame and vehicle account information corresponding to the three-dimensional detection frame of the target vehicle in each frame; wherein the target truth value frame of the three-dimensional detection frame in each frame is the same; Based on the posture transformation algorithm, the coordinate system of the three-dimensional detection frame in each frame is transformed to determine the transformed three-dimensional detection frame in each frame; Determine the detection track of the target vehicle based on the vehicle account information and the converted three-dimensional detection frame of each frame; Determine a difference between a center point of the converted three-dimensional detection frame in the first frame and a center point of the target true value frame in the first frame as a first side length; Determine the difference between the center point of the converted three-dimensional detection frame in the second frame and the center point of the target true value frame in the second frame as the second side length; Determine the displacement length of the true value trajectory between the first frame and the second frame and the displacement length of the detection trajectory as the third side length and the fourth side length respectively; Determine a geometric area formed by the first side length, the second side length, the third side length, and the fourth side length; Based on the geometric area in every two frames, the center point error area of ​​the target vehicle is determined to adjust the three-dimensional target detection algorithm according to the center point error area; the three-dimensional target detection algorithm is used to detect the three-dimensional detection frame of the target vehicle in each frame.

2. The determination method according to claim 1, characterized in that: The coordinate system transformation of the three-dimensional detection frame in each frame based on the posture transformation algorithm is performed to determine the transformed three-dimensional detection frame in each frame, including: Obtaining coordinates of eight vertices of the three-dimensional detection frame in the vehicle coordinate system; Based on the eight vertex coordinates of the three-dimensional detection frame in the vehicle coordinate system, the rotation matrix, and the translation projection coefficients from the vehicle coordinate system to the world coordinate system, the three-dimensional detection frame after conversion in each frame is determined.

3. The determination method according to claim 1, characterized in that: The step of determining the detection track of the target vehicle based on the vehicle account information and the converted three-dimensional detection frame of each frame includes: Based on the eight vertex coordinates of the three-dimensional detection frame after conversion in each frame in the world coordinate system, determining the center point coordinates of the three-dimensional detection frame after conversion in each frame in the world coordinate system; Based on the vehicle account information, the center point coordinates of the converted three-dimensional detection frame in each frame in the world coordinate system are connected to determine the detection trajectory of the target vehicle.

4. The determination method according to claim 1, characterized in that: The step of determining the center point error area of ​​the target vehicle based on the geometric area in every two frames includes: The geometric areas of every two frames are summed to determine the total geometric area; The total geometric area is averaged to determine the center point error area of ​​the target vehicle.

5. The determination method according to claim 1, characterized in that: After determining the detection track of the target vehicle based on the vehicle account information and the three-dimensional detection frame converted in each frame, the determination method further includes: Determine a heading angle difference between a heading angle of the three-dimensional detection frame in a world coordinate system corresponding to each frame and a heading angle of the target true value frame in the world coordinate system; The heading angle errors corresponding to each frame are summed to determine the total heading angle difference; Calculating the mean of the total heading angle differences to determine the heading angle error of the target vehicle; The three-dimensional target detection algorithm is adjusted based on the heading angle error area of ​​the target vehicle.

6. The determination method according to claim 1, characterized in that: The method of determining the target true value frame and vehicle account information corresponding to the three-dimensional detection frame of the target vehicle in each frame based on the pre-annotated multiple true value frames includes: Determine an intersection-over-union ratio between each of the true value frames and the three-dimensional detection frame of the target vehicle in each frame; The maximum intersection-and-union ratio is screened out from the multiple intersection-and-union ratios, the truth value frame corresponding to the maximum intersection-and-union ratio is determined as the target truth value frame corresponding to the three-dimensional detection frame, and the vehicle account information of the target truth value frame is determined as the vehicle account information corresponding to the three-dimensional detection frame.

7. A device for determining target detection evaluation indicators, characterized in that: The determining device comprises: A first determination module is used to determine a target truth frame and vehicle account information corresponding to a three-dimensional detection frame of a target vehicle in each frame based on a plurality of pre-labeled truth frames; wherein the target truth frame of the three-dimensional detection frame in each frame is the same; A posture transformation module is used to transform the coordinate system of the three-dimensional detection frame in each frame based on the posture transformation algorithm, and determine the transformed three-dimensional detection frame in each frame; A trajectory determination module, used to determine the detection trajectory of the target vehicle based on the vehicle account information and the converted three-dimensional detection frame of each frame; The first error determination module is used to determine the difference between the center point of the converted three-dimensional detection frame in the first frame and the center point of the target true value frame in the first frame as the first side length; determine the difference between the center point of the converted three-dimensional detection frame in the second frame and the center point of the target true value frame in the second frame as the second side length; determine the displacement length of the true value trajectory between the first frame and the second frame and the displacement length of the detection trajectory as the third side length and the fourth side length respectively; determine the geometric area formed by the first side length, the second side length, the third side length and the fourth side length; based on the geometric area under every two frames, determine the center point error area of ​​the target vehicle, so as to adjust the three-dimensional target detection algorithm according to the center point error area; the three-dimensional target detection algorithm is used to detect the three-dimensional detection frame of the target vehicle in each frame.

8. The determination device according to claim 7, characterized in that: When the posture transformation module is used to perform coordinate system transformation on the three-dimensional detection frame in each frame based on the posture transformation algorithm to determine the three-dimensional detection frame after transformation in each frame, the posture transformation module is specifically used to: Obtaining coordinates of eight vertices of the three-dimensional detection frame in the vehicle coordinate system; Based on the eight vertex coordinates of the three-dimensional detection frame in the vehicle coordinate system, the rotation matrix, and the translation projection coefficients from the vehicle coordinate system to the world coordinate system, the three-dimensional detection frame after conversion in each frame is determined.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and the machine-readable instructions are executed by the processor to execute the steps of the method for determining the target detection evaluation index as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for determining the target detection evaluation index according to any one of claims 1 to 6 are executed.