Research On The Technology Of Region Of Interest Prediction In Lane Detection

Lane detection based on machine vision is widely used in automotive active safety technology,which can effectively prevent the occurrence of traffic accidents,reducing casualties and economic losses.How to minimize the area of image processing in lane detection is the key to improve the real-time and anti-jamming ability of the system.After considering the relationship between vehicle motion and lane changing,this paper established the micro-traffic-system model,which is applied to predict region of interest by Kalman filter and deviation correcting method,and integrated into the lane detection system.The experimental results under various conditions show that the proposed method can effectively decrease the area of ROI,thus reducing the amount of image processing,and improving the real-time performance and anti-jamming level.The ROI prediction method based on video sequence without fully consider the characters of vehicle's motion will come to an inaccurate result when the lateral motion of the vehicle is increasing.Juggling accuracy and simplicity,this paper established the micro-traffic-system stem from vehicle dynamics with the freedom of two degree after considering the relationship between vehicle's motion and lane changing.By comparing the system model with commercial software,the applicability of the micro-trafficsystem model in the prediction of ROI is verified.Furthermore,as the interval between adjacent image frames is so short,this paper takes the longitudinal velocity of each frame as the input value in each period as approximation so as to discretize the system,getting the state space model with steering wheel and speed value as input.Based on this model,ROI prediction in vehicle coordinate system is realized by the method of deviation correcting using Kalman filter.In addition,the noise model,obtained by comparing the commercial software with the ideal model,is applied in simulation to test how much this method can reduce the area of interest.In order to verify the ROI prediction method proposed in this paper,lane detection system is developed by NI EVS.With the help of hardware in the loop test bench,the performance of the algorithm is tested with virtual condition generated by PreScan and experimental data from real car.The result shows that the ROI prediction based on vehicle dynamics model can not only effectively reduce the image processing area,but also improve the real-time performance and anti-jamming level of the system.