Aiming at the problem of large detection errors caused by the difficulty of lane feature extraction and target capture in variable illumination,this paper put forward a method of detecting lane lines based on improved dee-plabv3+.At first,a mathematical model of the lane was constructed in the form of small line segments.Corresponding-ly,the difference of distribution distance between points on the left and right lanes was calculated by the line fitting method.Then,the corresponding relationship of points on both sides was obtained after the difference comparison.Meanwhile,the maximum range of detection targets based on the threshold constraint was determined.Moreover,the image grid was built by the improved deeplabv3+algorithm.The four boundaries of the grid should have the same con-fidence.After that,the best threshold for lane detection was calculated,and the lane transformation parameters under the illumination environment were used as the contrast space threshold.Furthermore,the comparison between target data was carried out.Finally,the detection value with the highest accuracy was obtained by searching for the peak val-ue.Thus,the lane detection was completed.Experimental results show that the proposed method has high detection accuracy and low false detection rate,and can adapt to different variable illumination.
维普
万方数据