Research On Lane Detection And Vehicle Deviation Early Warning Based On Machine Vision

Advanced Driving Assistance System,which uses body hardware to sense the vehicle’s surroundings for data collection and systematic computation and analysis of vehicle data information,remind drivers of potential hazards and take corresponding measures to effectively improve the driving comfort and safety.Lane detection and vehicle deviation early warning technology as an important component of intelligent assistant driving system,how to achieve better lane detection and early warning based on the existing visual hardware has become a hot spot in the industry.In this paper,the road environment perception based on monocular camera is combined with the improved image processing algorithm to realize the improvement of the front lane recognition and vehicle departure warning function in the structured road,the main research work is as follows:(1)In order to ensure that the following lane detection has simplified data input,the road image is preprocessed.The lane-line image after distortion correction is gray processed by multi-channel weighted image,which makes the main lane-line and the background of the environment more distinct The speed of threshold segmentation is improved by improving the optimal threshold solution method of Otsu algorithm,and then the HED model is used to extract lane edge features and select the region of interest of Lane,an adaptive inverse perspective transform model is used to transform lane image into parallel mode.The results show that the lane features are obvious,the image noise is less and the processing speed is improved.(2)Conventional lane detection methods have the disadvantage of decreasing accuracy or real-time performance when fitting corners.Improved sliding window method combined with extended Kalman filter is used for lane detection and tracking.Taking the pre-processed image as the input of the algorithm,the starting position of the left and right lane pixels in the image is determined firstly,and then the starting position is taken as the initial window position and is sliding according to the rules,the window pixels are obtained and the lane curve is fitted.After the window traverses the image,the lane curve is smoothed and optimized,and the lane detection result is obtained Based on the above detection results,Lane tracking is combined with Kalman filter to improve the lane recognition rate in complex environment.Experimental results show that the improved lane detection algorithm can take into account both straight and curve recognition,and improve the accuracy of the whole lane detection while ensuring the real-time algorithm.(3)Based on the advantage of TLC Early Warning model in early warning time,the calculation process and early warning decision are optimized to improve the accuracy of early warning.Firstly,the exterior parameters of the camera are calculated based on the principle of camera imaging,and then the vehicle yaw angle is calculated based on the principle of TLC model,using Lane detection information,the lateral deviation distance of vehicles is improved and the curvature radius is calculated.According to the obtained parameters,the time required for vehicles to cross the lane is calculated Finally,the vehicle motion state is judged by the evaluation function,and the vehicle deviation early-warning output is carried out with the system threshold.Through experimental verification and analysis,the method in this paper performs well in various environments,can avoid some false alarms,and has a certain improvement in the accuracy of early warning,real-time performance can meet the system requirements.