Vision Based Vehicle External Environment Perception System

In the unmanned technology,vehicle environmental perception is an important part of the vehicle advanced assisted driving system(ADAS),which is the premise of control decision-making and the guarantee of safe driving.Visual-based external environment perception system is an effective means to realize ADAS because of its rich image information and relatively low cost.In this paper,panoramic view,lane detection and vehicle detection technologies are studied based on vision sensor in vehicle environment perception.The research contents and results are as follows:(1)A real-time inverse perspective transformation method proposed by wide-angle camera is used to realize the acquisition system of environmental information around vehicles.Firstly,the wide-angle camera is used to acquire real-time videos from the front,rear,right and the left of the vehicle.Then,the internal and external parameters of the camera are obtained by using the Zhang’s calibration method,and the wide-angle distorted image is corrected.By means of the inverse perspective transformation,the coordinate mapping relationship between the image coordinate and the world coordinate is established.The complex inverse perspective transformation formula is transformed into a look-up table mode which can realize real-time interpolation calculation,thus greatly improving the processing speed of the system.Finally,the omni-direction videos are stitched and displayed.The experimental results show that the algorithm can realize real-time image acquisition on DSP,and the acquisition frame rate can reach 22 frames per second.(2)A real-time lane recognition system is implemented based on monocular vision,which uses edge features to extract lane edge points,and uses least squares method to cluster the edge points.Firstly,image preprocessing includes image graying,median filtering and Sobel edge detection;Secondly,the maximum value of each region is found as the edge anchor point by using 5×5 raster scanning gradient image,and the image edge is connected according to the anchor point to reconstruct the lane line;finally,a method of processing edge points by least squares method is proposed for edge image.Each connected area is fitted by the least square method,and the edge points which do not conform to the lane line are excluded according to the characteristics that the left lane line is at a certain angle.Experimental results show that the proposed algorithm is faster than the Hough transform algorithm,and the linear fitting results of the two algorithms are consistent.(3)Based on monocular vision,a method of recognition of transmitting scan line is proposed to realize the real-time detection system of vehicle in front.Firstly,the region of interest(ROI)in front of the vehicle is determined according to the position of vanishing point and the transmission scanning line of the image;secondly,the gray threshold is calculated by means of mean variance method,and then the image is binarized according to the threshold to obtain the shadow target area of the vehicle bottom,i.e.the vehicle hypothesis area;finally,the hypothesis area is tested according to the characteristics of vehicle width,symmetry and information entropy.In order to improve robustness and accuracy,the error detection area is eliminated.The algorithm is transplanted to the DSP,and the real-time vehicle detection is realized on the DSP.The experimental results show that the algorithm can achieve real-time and stable detection of vehicles in front of the vehicle under different illumination intensity.