Evaluation method and evaluation system for lane boundary detection
The evaluation method and system use vehicle movement data to estimate lane boundaries, aligning frames with rotation and translation matrices, addressing the need for high-precision sensors and maps, thus providing cost-effective and accurate lane boundary detection.
Patent Information
- Application Number
- JP2025044788
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2024-03-21
- Filing Date
- 2025-03-19
- Publication Date
- 2025-10-03
AI Technical Summary
Existing methods for evaluating lane boundary detection in autonomous driving require high-precision sensors and maps, leading to high costs and complex processing, making them impractical for widespread implementation.
An evaluation method and system that utilizes vehicle movement data to estimate lane boundary lines by combining boundary envelope parameters with vehicle movement data, forming ground truth lines without the need for high-precision sensors or maps, using rotation and translation matrices to align frames and compare with detected lane boundaries.
Enables accurate evaluation of lane boundary detection without additional sensors or high-resolution maps, reducing costs and complexity while maintaining evaluation accuracy.
Smart Images
Figure 2025146785000001_ABST
Abstract
Description
[Technical Field]
[0001] The present invention relates to the field of intelligent driving technology, and in particular to an evaluation method and evaluation system for lane boundary detection. [Background technology]
[0002] Autonomous driving functions such as ACC (Adaptive Cruise Control) or LKS (Lane Keeping Assist System) are mature solutions on the market, and both require lane boundary information to help select target objects. In this context, it is important to evaluate whether lane boundary detection is accurate for software quality validation.
[0003] In particular, prior art methods for evaluating lane line detection typically use higher accuracy sensors and high resolution maps to obtain ground truth lane line information, an approach that requires additional, typically expensive, equipment and requires additional fusion algorithms.
[0004] Furthermore, another method is to train a large AI model to identify information about false positives and thereby evaluate lane boundary detection, but this requires complex processing methods and is difficult to achieve the desired results.
[0005] As mentioned above, prior art solutions often require high costs and more involvement from the developer. Summary of the Invention [Problem to be solved by the invention]
[0006] Based on the above-mentioned problems of the prior art, the present invention aims to provide an evaluation method and evaluation system for lane boundary line detection that can realize evaluation of lane boundary line detection without requiring a high-precision sensor and a high-precision map. [Means for solving the problem]
[0007] An evaluation method for detecting lane boundary lines according to one aspect of the present invention includes: acquiring vehicle movement data; extracting boundary envelope line parameters from left and right boundary points of the vehicle at a specified distance from the vehicle; Combining the vehicle movement data and the boundary envelope parameters in a corresponding format for each frame; storing the combined vehicle movement data and boundary envelope parameters in a designated file; Extracting, from a specified file, boundary envelope parameters of a specified frame after the current frame and vehicle movement data associated with the boundary envelope parameters; obtaining a rotation matrix and a translation variable based on the extracted vehicle movement data; Obtaining position information of a point in a specified frame after the current frame based on a boundary envelope parameter of the specified frame after the current frame is extracted; Obtaining position information of the point in a specified frame after the current frame using the rotation matrix and the translation variables, and rotating and translating the position information of the point in the current image; forming points in the current image as ground truth, and then further forming a ground truth line; evaluating the lane boundary line detection by comparing the difference between the ground truth line and the lane boundary line obtained by the lane boundary line detection; Includes.
[0008] An evaluation system for detecting lane boundary lines according to one aspect of the present invention includes: an acquisition module for acquiring vehicle movement data; an extraction module for extracting boundary envelope line parameters from left and right boundary points of the vehicle at a specified distance from the vehicle; a combining module for combining the vehicle movement data and the boundary envelope parameters in a corresponding format on a frame-by-frame basis; a storage module for storing the combined vehicle movement data and boundary envelope line parameters; a translation module for extracting boundary envelope line parameters and vehicle movement data associated with the boundary envelope line parameters of a specified frame after the current frame from a specified file, obtaining a rotation matrix and a translation variable according to the extracted vehicle movement data, obtaining position information of points of the specified frame after the current frame based on the boundary envelope parameters of the specified frame after the current frame is extracted, and using the rotation matrix and the translation variable to obtain the position information of points of the specified frame after the current frame, rotating and translating the position information of points in the current image, forming the points in the current image as a ground truth, and then further forming a ground truth line; a comparison module for comparing a difference between the ground truth line and the lane boundary line obtained by the lane boundary line detection to perform evaluation of the lane boundary line detection; Equipped with.
[0009] The foregoing and other objects and advantages of the present application will become more fully apparent from the following detailed description taken in conjunction with the accompanying drawings, in which like or similar elements are designated by like reference numerals. [Brief explanation of the drawings]
[0010] [Figure 1] 3 is a flowchart of an evaluation method for lane boundary line detection according to an embodiment of the present invention. [Figure 2] FIG. 2 is a visual schematic diagram of an evaluation method for lane boundary detection according to an embodiment of the present invention. [Figure 3] 1 is a structural diagram of an evaluation system for lane boundary line detection according to an embodiment of the present invention; DETAILED DESCRIPTION OF THE INVENTION
[0011] The following describes some embodiments of the present disclosure and is intended to provide a basic understanding of the disclosure, but is not intended to identify key or critical elements of the invention or to limit the scope of protection sought.
[0012] First, the evaluation method for lane boundary line detection according to the present invention mainly assumes that future consecutive frames are used to estimate lane boundary lines of the current frame, and if the nearby lane boundary lines are detected well enough, they can be considered as the ground truth. Based on this assumption, the present invention takes lane boundary line points at a specified distance (e.g., preferably 8 meters) from the vehicle, combines the points of several frames into the current frame, and uses the result as the ground truth for the current frame.
[0013] FIG. 1 is a flowchart of an evaluation method for lane boundary line detection according to an embodiment of the present invention.
[0014] Step S1: Read vehicle movement data from the mf4 file, which includes at least the yaw angle and vehicle speed (the yaw angle and vehicle speed are used to subsequently calculate the rotation matrix T and translation vector t).
[0015] Step S2: The horizontal deviation, heading angle, curvature, and curvature change of the vehicle boundary point at a specific distance of the vehicle are extracted. These parameters refer to the step number of the lane boundary line where the sampling point is located. As a preferred method, the vehicle boundary point at a distance of 8 meters is extracted because 8 meters is the vertical distance under the world coordinate system corresponding to the bottom point of the image under the camera FOV viewing angle.
[0016] This is the closest distance point that the camera can see. This is the preferred method, and a point 8 meters from the vehicle is used, because the closer the road is to the camera, the less the effect of road slope will be and the more accurate the estimated position in the world coordinate system will be.
[0017] Step S3: The vehicle movement data and boundary envelope line parameters are bounded according to each frame, i.e., the points extracted from each frame are bounded one by one by the vehicle movement data under the corresponding frame number. The boundary envelope line parameters refer to the coefficients of each order of the boundary lane boundary line, including the horizontal deviation at the zero point, the angle between the zero points, the curvature at the zero point, and the rate of change of the curvature. Because the boundary envelope line parameters are already a curve equation described in the world coordinate system, the position information of the points can be obtained using the curve equation, i.e., the horizontal coordinates of the points at different predetermined longitudinal distances (e.g., 10 m, 15 m, 20 m, 25 m, etc.). Meanwhile, since the speed and yaw angle change in real time, the vehicle movement data (i.e., the speed and yaw angle) of each frame are different. Therefore, the position information of the points under each frame is combined with the vehicle movement data one by one, for example, combining the position information of the points in frame 1 with the vehicle movement data of frame 1, combining the position information of the points in frame 2 with the vehicle movement data of frame 2, etc.
[0018] Step S4: Export the combined data into a CSV (Comma-Separated Values) file.
[0019] Step S5: Simultaneously extract the point position information (including the ordinate and abscissa of the world coordinates) of the designated frame after the current frame and the vehicle movement data combined with the point information (i.e., yaw angle and velocity). Here, the "designated frame" can be an appropriate frame number without too many or too few frames. This is because the more frames selected, the longer the ground truth line will be. However, assuming the data has a total of 1,000 frames, if 500 consecutive frames are used to extract point information and obtain the ground truth line, the frames from which the ground truth line can be obtained can be obtained in the first 1, 2, 3, ... 500 frames. However, for frames 501 and larger, the remaining frame numbers are less than 500 frames, so the ground truth line cannot be obtained. Conversely, if there are too few frames, the ground truth value line will be too short. Here, as an example, point information in 50 frames is extracted.
[0020] Step S6: Using the rotation matrix R and translation vector t, rotate the dots of the specified frame extracted after the current frame onto the current image. The rotation matrix R and translation vector t are obtained according to the yaw angle and vehicle speed. In particular, the dots recorded 8 meters ahead of each frame are based on the vehicle's rear axis coordinate system for each frame. Because the vehicle is in a moving state, including translational and rotational movements, the coordinate system changes in real time. Now, assume that we want to determine the position of the dots recorded in each frame n frames below the nth frame. This requires rotational and translational operations. The dots from the (n+1)th frame projected onto the nth frame are given by p(n+1)·R(n)+t(n)=p(n+1)(n). The dots from the (n+2)th frame projected onto the nth frame are given by (p(n+2)·R(n+1)+t(n+1))·R(n)+t(n)=p(n+2)(n). The point from the (n+3)th frame projected onto the nth frame is ((p(n+3)·R(n+2)+t(n+2))·R(n+1)+t(n+1))·R(n)+t(n)=p(n+3)(n), etc.
[0021] where R represents the rotation matrix, t represents the translation vector, and p represents the world coordinates of the points, i.e., the second point requires one rotation and translation operation (from the first point to the second point), and the third point requires two such operations (from the first point to the second point, and from the second point to the third point).
[0022] Step S7: Form a ground truth value line from the transformed points, i.e., the ground truth value points.
[0023] Step S8: Visually display the current lane boundary line and the lane boundary line detection output by comparing the true line and the current lane boundary line. If the horizontal deviation of the two lines at the same longitudinal distance exceeds a threshold, the lane boundary line detection output for the current lane boundary line is evaluated as inaccurate.
[0024] Among them, in step S5, the specific form for obtaining the rotation matrix R and the translation vector t according to the yaw angle and the vehicle velocity is as follows:
[0025] The rotation matrix R is obtained by the following formula:
number
[0026] The translation vector t is obtained by the following equation:
number
[0027] In the above process, as an example, steps S1 to S4 can be performed via Python.
[0028] Steps S5 to S8 can be implemented by using C++ embedded in a simulation tool.
[0029] FIG. 2 is a visual schematic diagram of an evaluation method for lane boundary detection according to an embodiment of the present invention.
[0030] The solid line indicated by gt in Figure 2 is the ground truth value line obtained by the evaluation method for lane boundary line detection shown in Figure 1. The dashed line indicated by P is the lane boundary line detected by the lane boundary line detection output. If the horizontal deviation between the two lines at the same longitudinal distance exceeds a threshold, the lane boundary line detected by the lane boundary line detection output is evaluated as inaccurate. From Figure 2, it can be seen that there is a distance between the ground truth value line and the horizontal deviation between the lane boundary lines output by the lane boundary line detection at the same longitudinal distance.
[0031] FIG. 3 is a structural diagram of an evaluation system for lane boundary line detection according to an embodiment of the present invention.
[0032] As shown in FIG. 3, the lane boundary line detection evaluation system 100 according to the present invention includes: an acquisition module 110 for acquiring vehicle movement data; an extraction module 120 for extracting boundary envelope line parameters from left and right boundary points of the vehicle at a specified distance from the vehicle; a combining module 130 for combining the vehicle movement data and the boundary envelope line parameters according to each frame; a storage module 140 for storing the combined vehicle movement data and boundary envelope line parameters; a translation module 150 for extracting boundary envelope line parameters and vehicle movement data associated with the boundary envelope line parameters of a specified frame after the current frame from a specified file, obtaining a rotation matrix and a translation variable according to the extracted vehicle movement data, obtaining position information of points of the specified frame after the current frame based on the boundary envelope parameters of the specified frame after the current frame is extracted, and using the rotation matrix and the translation variable to obtain the position information of points of the specified frame after the current frame, rotating and translating the position information of points in the current image, forming the points in the current image as a ground truth, and then further forming a ground truth line; a comparison module 160 for comparing the difference between the ground truth line and the lane line obtained by the lane line detection to perform evaluation of the lane line detection; Equipped with.
[0033] Vehicle movement data includes yaw angle and vehicle speed, and boundary envelope line parameters include horizontal deviation, heading angle, curvature, and curvature change.
[0034] The comparison module 160 compares the ground truth value line with the vehicle lane line to detect horizontal deviations at the same longitudinal distance of the lane line output, and if the horizontal deviation exceeds a specified predetermined threshold, the lane line output of the vehicle lane line detection is evaluated as inaccurate.
[0035] The distance to the vehicle is preferably at a specified distance of 8 meters from the vehicle.
[0036] The present invention also provides a computer readable medium having stored thereon a computer program which, when executed by a processor, performs the evaluation method for lane boundary line detection.
[0037] The present disclosure also provides a computer device comprising a storage module, a processor, and a computer program stored on the storage module and executable on the processor, the computer device implementing the above evaluation method for lane boundary line detection when the processor executes the computer program.
[0038] According to the method and system for evaluating lane boundary line detection of the present invention, the output quality of lane boundary lines of a lane boundary line detection can be evaluated without additional sensors and high resolution maps.
[0039] The above description is only about specific embodiments of the present application, and the scope of protection of the present application is not limited thereto. Those skilled in the art may contemplate other feasible modifications or substitutions according to the technical scope disclosed in the present application, and all such modifications or substitutions fall within the scope of protection of the present application. Where there is no contradiction, the embodiments of the present application and the features within the embodiments can also be combined with each other. The scope of protection of the present application is subject to the scope of protection of the claims.
Claims
1. 1. An evaluation method for lane boundary detection, comprising: acquiring vehicle movement data; extracting boundary envelope line parameters from left and right boundary points of the vehicle at a specified distance from the vehicle; combining the vehicle movement data and the boundary envelope parameters in a corresponding format for each frame; storing the combined vehicle movement data and the boundary envelope parameters in a designated file; extracting, from the specified file, the boundary envelope parameters of a specified frame after the current frame and the vehicle movement data associated with the boundary envelope parameters; obtaining a rotation matrix and a translation variable based on the extracted vehicle movement data; Obtaining position information of a point of a designated frame after the current frame based on a boundary envelope parameter of the designated frame after the current frame is extracted; using the rotation matrix and the translation variables to obtain position information of points in a specified frame after the current frame, and rotating and translating the position information of points in the current image; forming points in the current image as ground truth, and then further forming a ground truth line; evaluating the lane boundary line detection by comparing a difference between the ground truth line and a lane boundary line obtained by lane boundary line detection; An evaluation method for lane boundary detection, including:
2. 2. The evaluation method for detecting lane boundary lines according to claim 1, wherein the vehicle movement data includes a yaw angle and a vehicle speed.
3. 2. The evaluation method for lane boundary line detection according to claim 1, wherein the boundary envelope line parameters include horizontal deviation, heading angle, curvature, and curvature change.
4. 2. The evaluation method for lane boundary line detection according to claim 1, wherein the evaluation of the lane boundary line detection by comparing the difference between the ground truth line and the lane boundary line obtained by the lane boundary line detection includes comparing the horizontal deviation between the ground truth line and the lane boundary line obtained by the lane boundary line detection at positions spaced apart in the same longitudinal direction, and if the horizontal deviation exceeds a predetermined threshold, the lane boundary line of the lane boundary line detection output of the vehicle is evaluated to be erroneous.
5. 2. The evaluation method for lane boundary line detection according to claim 1, wherein the distance is a specified distance of the vehicle at a distance of 8 meters from the vehicle.
6. 1. An evaluation system for lane boundary detection, comprising: an acquisition module for acquiring vehicle movement data; an extraction module for extracting boundary envelope line parameters from left and right boundary points of the vehicle at a specified distance from the vehicle; a combining module for combining the vehicle movement data and the boundary envelope parameters in a corresponding manner on a frame-by-frame basis; a storage module for storing the combined vehicle movement data and the boundary envelope line parameters; a translation module for extracting the boundary envelope line parameters and the vehicle movement data combined with the boundary envelope line parameters of a designated frame after the current frame from a designated file, obtaining a rotation matrix and a translation variable based on the extracted vehicle movement data, obtaining position information of points of the designated frame after the current frame based on the boundary envelope parameters of the designated frame after the current frame is extracted, using the rotation matrix and the translation variable to obtain position information of points of the designated frame after the current frame, rotating and translating the position information of points in the current image, forming the points in the current image as a ground truth, and then further forming a ground truth line; a comparison module for comparing a difference between the ground truth line and a lane boundary line obtained by lane boundary line detection to perform evaluation of the lane boundary line detection; An evaluation system for lane boundary line detection, comprising:
7. 7. The evaluation system for lane boundary line detection according to claim 6, wherein the vehicle movement data includes yaw angle and vehicle speed.
8. 7. The evaluation system for lane boundary line detection according to claim 6, wherein the boundary envelope line parameters include horizontal deviation, heading angle, curvature, and curvature change.
9. 2. The evaluation system for lane boundary line detection according to claim 1, wherein the comparison module compares the horizontal deviation of the lane boundary line of the ground truth line with the vehicle boundary line detection output at the same longitudinal distance, and if the horizontal deviation exceeds a predetermined threshold, the lane boundary line of the vehicle boundary line detection output is evaluated as inaccurate.
10. 7. The evaluation system for lane line detection of claim 6, wherein the specified distance from the vehicle is 8 meters from the vehicle.
11. 6. A computer-readable medium storing a computer program, the computer program being adapted to implement the evaluation method for lane boundary line detection according to any one of claims 1 to 5 when executed by a processor.
12. 1. A computer device comprising: a storage module; a processor; and a computer program stored on the storage module and executable on the processor, A computer device that performs the evaluation method for detecting lane boundary lines according to any one of claims 1 to 5 when the processor executes the computer program.
13. 6. A computer program product comprising a computer program that, when executed by a processor, performs the method for detecting and evaluating lane boundary lines according to any one of claims 1 to 5.