Research On Autonomous Vehicle Road Detection Algorithm Based On Machine Vision

Unmanned driving has always been a hot issue of research.Unmanned road detection technology is the key to ensuring the safe driving of unmanned vehicles.Road detection technology is one of the important research directions of unmanned driving.Its main function is to identify lane lines and road boundaries,and delimit the driving area of unmanned vehicles on the road,so as to ensure the driving safety of unmanned vehicles.This article first preprocesses the road image,uses camera calibration,determines the region of interest,processes the image by methods such as perspective transformation and image gray-scale,so as to create conditions for the next step of road detection.In view of the difference between structured roads and unstructured roads,different detection methods are proposed.The lane lines of structured roads are generally more obvious.After preprocessing the image,by comparing the detection results of the Sobel edge detection operator,a scheme is proposed to detect yellow lane lines and white lane lines separately.The RGB channels are converted into HSL model and Lab model,the L channel in the HSL model is used to detect the white lane line in the image,and the b channel in the Lab model is used to detect the yellow lane line in the image.The pixel coordinates of the lane line are determined by the sliding window polynomial fitting method,and finally the curvature of the lane line and the relative offset distance of the vehicle are calculated and projected into the image.Non-institutional roads usually do not have obvious lane lines,but have road boundaries.Through the method of image semantic segmentation.And combined with the neural network to complete the detection.By comparing the structure and characteristics of various neural networks,the FCN full convolutional network is used to segment the road,and the semantic understanding of the road scene is realized.Adjust the corresponding value of the neural network unit to prevent over-fitting,so as to ensure a good segmentation effect.