Vehicle Detection And Tracking Research Based On Low-altitude Platform

Machine vision has got a great achievement in intelligent traffic supervision system as it develops nowadays. It also has more thorough research in vehicle detecting and tracking. Among these subjects vehicle detection and tracking on Low-altitude platform is a challenge.This paper analyses video sequences of traffic intersection on low-altitude platform. After studying the algorithms about target detection and tracking, we present a new method using detecting and tracking algorithm to construct the road supervision system, at the same time analyze road information. Our work can be presented as following:(1) Making research on vehicle video sequences of traffic intersection. We did experiments about traditional frame-difference method and background-reduction method. As it didn't show ideal effect we use a new method combing difference-accumulation and Gauss background modeling to extract vehicles running on the road. Experimental results show it has good performances.(2) Foreground sequences always have noises. We calculate the threshold of the noise area after sufficient experiments, then remove the isolated noise areas according to the threshold in order that foreground sequences are more accurate. At last do preparation for the following tracking.(3) Due to the interferences of illumination and shadow, the detected rectangles are usually bigger than the figures of the target areas. We educe the proportion of detected area and real vehicle through abundant experiments, set the proportion 1:0.8, zoom the detected areas which are used to be tracked.(4) In this paper, we compared results of Mean-shift,Camshift,ABCshift,Kalman and Particle Filter in low-altitude platform. We use a Particle Filter method based on Mean-shift to solve the defects of these five methods, reduce the number of particles that track the target meanwhile increase speed by optimize particles.(5) This paper integrates detection and tracking, alternates between detecting and tracking.(6) The system uses Visual c++ as tools to develop algorithms, choose OpenCV image lib to collect video sequences and realize some algorithms. Experiments prove this method is valid.