Monocular Vision Based Road Vehicle Detection

Vehicle detection is not only one of the most significant fields in road environment perception technology.It is also widely used for adaptive cruise control,automatic emergency braking and other advanced driving assistant systems.Thus,a research about vision-based vehicle identification technique is carried out on accounts of its widespread application prospect.In order to reduce the computing consumption,a search space reduced method is proposed.At first,a pinhole camera model is used for calculating the position of vanishing line precisely.With the vanishing line,the sky part of picture is split out and a reduced initial detection region is generated.Further,road is also quickly and precisely divided for reducing the vehicle detection area by the lane marker.Along with Canny method is selected to extract edge points,Hough transform is selected to find out the straight line in the picture.To ensure the edge extraction,an adaptive Canny thresholds method based on the gray gradient distribution is proposed.To overcome the time-consuming problem,the area of road edge points and vote space of Hough transform is restrained with road vanishing point,the width of road.In addition,in order to ensure the accuracy of the road vanishing point,its location is updated in real-time by clustering road line meet points.In order to detect the vehicle precisely and steadily,two vehicle classifiers is trained by HOG feature and SVM method.They all achieve 97% accuracy result.A rapid vehicle detection method based on shadows is used in this paper.Considering the problem that road surface may be under different illumination,a novel approach using gray level distribution and space correlation is proposed to split sub-road area.And the inaccuracy of locating based on single shadow feature is overcome by using the left,right and upper edge feature of the vehicle.In this paper vehicle detection with sliding windows method was also discussed and a knowledge-based image pyramid is proposed to reduce computing consumption.It used camera model and the width of road to reduce the size of pictures in the pyramid image.At last,tracking target and the location of target is predicted by Kalman filter when the target is not detected temporarily.With the supplement of predicted location,vehicle detection result can be improved.