Structured Curve Lane Detection And Its Embedded Implemention

The main work of this paper has focused on the detection of structured curve lane.The key research is about two methods that one is based on the three-segment and the other is based on the birdview edge,and the implementation in the embedded platform of the method based on the three-segment.The main work and innovation of this paper are as follows:1.Sobel operator,Canny operator and PNG algorithm are compared and analyzed,and an improved method is proposed for PNG algorithm.The basic idea of the PNG algorithm is to match the left and right edges in the image.The gradient amplitude histogram is used to determine the edge threshold,which can solve the problem of missing edge information caused by fixed threshold and avoid too much noise.By using the left edge search range associated with the lane line width,the resulting edge points are more physically meaningful and the generation of interference edges is avoided.The edge feature is the foundation of lane detection,and the improved PNG algorithm can get the edge feature to create a good foundation for the following work of lane detection.2.A curve lane detection algorithm based on three-segment is proposed.The segmentation of the processed image region is completed by the segmentation of the forward distance of the world coordinate system.The Hough transform based on the vanishing point is respectively performed on the three segments,and then the segment line detection results are obtained through three segments respectively.The results of the three-segment line results and their reliability are used to fit the hyperbolic model of lane.Experiments show that the algorithm has better reliability and it has better description of curve lane than the straight line model.3.A curve lane detection algorithm based on birdview edge is proposed.This method is to find the lane line model in a small range of the previous frame detection results,and to project the edge to the birdview by inverse perspective transformation.According to the lane line detection results,the edge points in the birdview which satisfy the curvature radius constraints and distance constraints are fitted to obtain the polynomial model parameters.4.The algorithm of curve detection based on three-segment is implemented on AP SoC embedded platform.According to the computation characteristics of the plarform,the edge detection and Hough transform are performed on the FPGA and the segment lane lines selection and the hyperbola model fitting are carried out on the ARM.After the offline test on a large number of real road environment images,the algorithm has been proved of strong accuracy.The frame rate can reach 15 frames per second,which meets the real-time need.