Target Detecting And Tracking Of Intelligent Vehicle On Structural Road

In recent years, due to the development of road transport and increasing in car ownership,road traffic safety issues have become prominent in our country. With the development of domestic and foreign unmanned technology, the emergence of intelligent vehicle provides the possibility of a substantial increase in automobile driving safety, improvement of the efficiency on road traffic and energy saving. As an important link of environmental perception part of intelligent vehicle, target detecting and tracking has also become one of the hotspots on research.The purpose is through detecting the car in front of the intelligent vehicle on the road to get driving area for intelligent vehicle and avoid collisions of vehicles.This paper make a depth research on vehicle detecting and vehicle tracking method, and effectively achieve vehicle detecting and tracking on structural road in the daytime. The main research work includes:(1) Proposed an improved sub-window adaptive threshold binarization method, extracting the lane line and establishing road detection region. Firstly, according to the projection relationship of the camera, the whole image is divided into numbers of different sizes of rectangular windows. Then, binarizing with Otsu in each of the rectangular window. Finally,measuring the complexity of the window by the contrast in grayscale of rectangular window,getting the window which contains the lane line and establishing road detection region. This method can avoid interference effectively to the lane line extraction, such as road shadows,smudges and water etc. The size of window is consistent with deformation of the road image,thus, effectively reducing producing of undesirable results.(2) Proposed a vehicle detecting method which based on multi-feature fusion. Firstly,according to the difference of gray value and gradient value between vehicle bottom shadow and road surface detection region, detecting vehicle bottom shadow. And generating hypothesis vehicle by vehicle bottom shadow. Then, extracting texture features of vehicle through calculating of fractal dimension, getting shape features of the vehicle by matching four edges of the vehicle with edge templates, obtaining symmetry features of the vehicle by measuring symmetry with projection of vehicle vertical edge. Finally, using geometric criteria which based on the variance measurement to establishing a vehicle judgment formula which fusion vehicle texture features, the shape feature and the vertical edge symmetry features. And using this vehicle judgment formula to verifying generate hypotheses vehicle. This approach has good adaptability under different lighting and road conditions. The average detection rate is 95.7%, the average false detection rate is 2.5%, the average missing detection rate is 1.8%, each frame detection time is around 47 ms.(3) Researched a vehicle tracking method based Kalman filter. Firstly, using location, width,height as the tracking component, and using Kalman filter predicting the position of vehicle.Then, using the four edges of the vehicle and the vehicle bottom shadow feature to complete relocation of target and achieve update of observation vector in the forecast area. Finally,determining vehicle based on normalized moment of inertia and information entropy. The results show that vehicle tracking algorithm can keep track of targets within 120 m range on road surface area, the average number of frames for each tracking target is more than 300.