Large-curvature lane line processing method and system based on Bezier curve

Abstract

The invention discloses a large-curvature lane line processing method and system based on a Bezier curve, and the method comprises the steps: collecting images of roads around a vehicle in real time in the vehicle driving process, directly obtaining a lane line point set under a vehicle body coordinate system through a bev network, carrying out the cubic fitting of the obtained point set, obtaining the curvature of the lane line, and carrying out the calculation of the curvature of the lane line. And if the curvature is greater than a preset threshold value, performing Bessel fitting on the obtained point set for three times to obtain a control point of a single frame, and tracking the control point. According to the method, the problem of large cubic fitting error in a large-curvature curve scene is effectively solved. The lane line plays an important role in the functions of automatic driving LCC, LKA and the like, and whether the vehicle can pass through the curve stably and smoothly or not can be directly determined according to the fitting precision. Meanwhile, compared with Kalman filtering, the control point tracking method provided by the invention has the advantages that the operation is more efficient, and the implementation is simpler.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a method and system for processing large curvature lane lines based on Bezier curves. Background Art

[0002] At present, there are many ways to process and track lane lines. Most of them use cubic polynomials to fit the lane line point set obtained by model detection, and then use the cubic coefficients of the lane line to mathematically model and use the Kalman filter to perform parameter filtering to achieve smoothing. However, cubic polynomials can only fit straight lines or curves with small curvatures. If the curvature is large, the cubic fitting cannot well reflect the actual curvature of the lane line and there is a large error. Summary of the invention

[0003] The purpose of the present invention is to provide a method and system for processing lane lines with large curvature based on Bezier curves, which can solve the problem of large cubic curve fitting errors under large curvatures, and at the same time perform multi-frame smoothing of lane lines by tracking control points. Compared with Kalman filtering, the calculation is more efficient and the implementation is simpler.

[0004] The present application provides a method for processing a large curvature lane line based on a Bezier curve, the method comprising the following steps: During the driving process of the vehicle, the on-board camera is used to collect real-time images of the road around the vehicle, and the BEV network is used to directly obtain the lane line point set in the vehicle body coordinate system. The obtained point set is fitted three times to obtain the curvature of the lane line. If the curvature is greater than the preset threshold, the obtained point set is fitted three times to obtain the control points of a single frame, and the control points are tracked.

[0005] As a further technical solution of the present invention, when the curvature is greater than a preset threshold, three Bezier fittings are performed on the obtained point set to obtain four control points.

[0006] As a further technical solution of the present invention, when fitting the obtained point set using the Bezier curve, the RANSAC algorithm is used to process outliers to obtain the optimal control points.

[0007] As a further technical solution of the present invention, the step of performing a cubic Bezier fitting on the obtained point set includes: Iterate multiple times, randomly select multiple points from the lane line point set each time, and the number of points randomly selected each time is the same, and calculate the control points of the set of points; Generate n points from each control point, and use these n points to calculate the error with the original n points as the evaluation index of the group of control points; The set of control points with the smallest error is selected as the optimal control points, which is the calculation result of the control points of a single frame.

[0008] As a further technical solution of the present invention, the error calculation method is: calculating the Euclidean distance of the point corresponding to the index, and obtaining the average value of the Euclidean distance of n points as the error judgment standard.

[0009] As a further technical solution of the present invention, the tracking of the control points refers to performing tracking filtering processing on the control points, including the following steps: Calculate the average Euclidean distance between the point set generated by the control point and the tracker's historical point set as the loss matrix value, and use Hungarian matching to solve the optimal matching; For the matched lane lines, the corresponding control points are filtered using alpha filtering, and the x-coordinate and y-coordinate values ​​of the control point corresponding to the index are respectively calculated using alpha filtering to obtain the final filtering results. The calculated average Euclidean distance must be less than a certain distance threshold, otherwise it is considered unmatched. The control points obtained after filtering need to be sampled to generate a smooth lane line point set as the final output.

[0010] Another object of the present invention is to provide a large curvature lane line processing system based on Bezier curves, the system comprising: An image acquisition module, used to collect images of the road surrounding the vehicle; Neural network reasoning module, used to obtain the lane line point set in the vehicle coordinate system; The curvature judgment module preliminarily fits the curvature of the lane line to determine whether it is a lane line with large curvature; Data processing module, used for lane line processing and control point tracking based on Bezier curves.

[0011] Beneficial effects achieved by the present invention: The lane line processing method and system based on Bezier curves proposed in the invention can effectively solve the problem of large cubic fitting errors in large curvature curves. Lane lines play an important role in autonomous driving functions such as LCC and LKA. The accuracy of fitting can directly determine whether the vehicle can smoothly pass the curve. At the same time, the control point tracking method proposed in the present invention is more efficient and simpler to implement than Kalman filtering. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a flow chart of a method for target detection based on a 4D millimeter wave radar sensor of the present invention.

[0013] Figure 2 This is a diagram of the Bezier curve fitting process of an embodiment of the present invention.

[0014] Figure 3 This is a control point tracking diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The technical solution of the present invention is described in detail below in conjunction with the specific drawings.

[0016] like Figure 1-3 As shown, the present application provides a method for processing a lane line with large curvature based on a Bezier curve, and the method comprises the following steps: During vehicle driving, an on-board camera is used to collect images of the road around the vehicle in real time, and the BEV network is used to directly obtain the lane line point set in the vehicle body coordinate system. The obtained point set is fitted three times to obtain the curvature of the lane line. If the curvature is greater than the preset threshold, the obtained point set is fitted three times to obtain the control points of a single frame, and the control points are tracked; if the curvature is not greater than the preset threshold, it can be processed according to the conventional method of the lane line. Please refer to the existing technology and will not be described in detail here.

[0017] In this embodiment, when the curvature is greater than a preset threshold, Bezier fitting is performed three times on the obtained point set to obtain four control points.

[0018] In this embodiment, when the obtained point set is fitted using the Bezier curve, the RANSAC algorithm is used to process outliers to obtain the optimal control points.

[0019] In this embodiment, the step of performing cubic Bezier fitting on the obtained point set includes: Iterate multiple times, randomly select multiple points from the lane line point set each time, and the number of points randomly selected each time is the same, and calculate the control points of the set of points; Generate n points from each control point, and use these n points to calculate the error with the original n points as the evaluation index of the group of control points; The set of control points with the smallest error is selected as the optimal control points, which is the calculation result of the control points of a single frame.

[0020] In this embodiment, the error calculation method is: calculate the Euclidean distance of the point corresponding to the index, and obtain the average value of the Euclidean distance of n points as the error judgment standard.

[0021] In this embodiment, the result of a single frame may fluctuate to a certain extent, so it is necessary to track the control points. The tracking of the control points refers to tracking and filtering the control points, including the following steps: Calculate the average Euclidean distance between the point set generated by the control points and the tracker's historical point set as the loss matrix value, and use Hungarian matching to solve the optimal matching. The tracker's historical point set refers to the point set generated by the control points in the tracker predicted to the current frame based on the vehicle posture information, that is, the position of the control points in the tracker is predicted based on the vehicle posture information, and the point set is generated using the control points predicted to the current frame; For the matched lane lines, the corresponding control points are filtered using alpha filtering, and the x-coordinate and y-coordinate values ​​of the control point corresponding to the index are respectively calculated using alpha filtering to obtain the final filtering results. The calculated average Euclidean distance must be less than a certain distance threshold, otherwise it is considered unmatched. The control points obtained after filtering need to be sampled to generate a smooth lane line point set as the final output.

[0022] Another object of the present invention is to provide a large curvature lane line processing system based on Bezier curves, the system comprising: An image acquisition module, used to collect images of the road surrounding the vehicle; Neural network reasoning module, used to obtain the lane line point set in the vehicle coordinate system; The curvature judgment module preliminarily fits the curvature of the lane line to determine whether it is a lane line with large curvature; Data processing module, used for lane line processing and control point tracking based on Bezier curves.

[0023] The image acquisition module, the neural network reasoning module, and the curvature judgment module are connected to the data processing module, and the large curvature lane line processing system based on the Bezier curve is configured to execute the steps of the above-mentioned large curvature lane line processing method based on the Bezier curve.

[0024] It should be noted that, in this article, the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0025] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for processing lane lines with large curvature based on Bezier curves, characterized in that: The method comprises the following steps: During the driving process of the vehicle, the road image around the vehicle is collected in real time, and the lane line point set in the vehicle body coordinate system is directly obtained using the BEV network. The obtained point set is fitted three times to obtain the curvature of the lane line. If the curvature is greater than the preset threshold, the obtained point set is fitted three times to obtain the control points of a single frame, and the control points are tracked.

2. The method for processing lane lines with large curvature based on Bezier curves according to claim 1, characterized in that: When the curvature is greater than a preset threshold, three Bezier fittings are performed on the obtained point set to obtain four control points.

3. The method for processing lane lines with large curvature based on Bezier curves according to claim 2, characterized in that: When fitting the obtained point set using the Bezier curve, the RANSAC algorithm is used to process outliers to obtain the optimal control points.

4. The method for processing lane lines with large curvature based on Bezier curves according to claim 3, characterized in that: The step of performing cubic Bezier fitting on the obtained point set comprises: Iterate multiple times, randomly select multiple points from the lane line point set each time, and the number of points randomly selected each time is the same, and calculate the control points of the set of points; Generate n points from each control point, and use these n points to calculate the error with the original n points as the evaluation index of the group of control points; The set of control points with the smallest error is selected as the optimal control points, which is the calculation result of the control points of a single frame.

5. The method for processing lane lines with large curvature based on Bezier curves according to claim 4, characterized in that: Error calculation method: Calculate the Euclidean distance of the point corresponding to the index, and take the average value of the Euclidean distance of n points as the error judgment criterion.

6. The method for processing lane lines with large curvature based on Bezier curves according to claim 1, characterized in that: Tracking the control points refers to performing tracking filtering on the control points, including the following steps: Calculate the average Euclidean distance between the point set generated by the control point and the tracker's historical point set as the loss matrix value, and use Hungarian matching to solve the optimal matching; For the matched lane lines, the corresponding control points are filtered using alpha filtering, and the x-coordinate and y-coordinate values ​​of the control point corresponding to the index are respectively calculated using alpha filtering to obtain the final filtering results. The calculated average Euclidean distance must be less than a certain distance threshold, otherwise it is considered unmatched. The control points obtained after filtering need to be sampled to generate a smooth lane line point set as the final output.

7. A large curvature lane line processing system based on Bezier curve, characterized in that: The system comprises: An image acquisition module, used to collect images of the road surrounding the vehicle; Neural network reasoning module, used to obtain the lane line point set in the vehicle coordinate system; The curvature judgment module preliminarily fits the curvature of the lane line to determine whether it is a lane line with large curvature; Data processing module, used for lane line processing and control point tracking based on Bezier curves.

Images