Video Based Vehicle Detection And Tracking Algorithms Research

In order to solve various problems caused by the rapid growing of surface traffic, the research of Intelligent Transportation System comes to an issue of consequence. As an important part of ITS, vehicle detection and tracking becomes a research focus all over the world. According to the key techniques of ITS, some algorithms and techniques for vehicle detection and tracking have been analyzed and improved.Several methods based on vehicle tracking, which are used to detect traffic incidents such as converse and over-speed, have been described in detail in this thesis. Moreover, a vehicle tracking and traffic incidents detection system has been designed for application. The main ideas of this thesis can be described as follow:1. During the algorithms of vehicle detection, the adaptive-k Gaussian mixture model is used for background rebuilding. Through the model building for each pixel, foreground mask could be got rightly. In the process of vehicle extraction, the accuracy and robustness of vehicle extraction could be enhanced since the vehicle extraction algorithms used is based on regional search algorithm.2. During the algorithm of vehicle tracking, this paper is multi-feature matching based on region algorithm. In the process of vehicle tracking, the features of size, shape, motion estimation and center position could be used in the vehicle tracking for the movement of vehicles in two adjacent mobile in the image change slowly. This ways can deal with less parameter and a small amount of data, so it could be used to track vehicles in the real time condition.3. Based on the result of vehicle tracking, this paper analysis some traffic incidents, such as the direction and speeding. Besides, proposed specific detection method and the experimental results are presented. For the vehicle speed detection, this paper discusses the camera calibration and the reasons for the errors.To sum up, according to some difficult problems in video vehicle detection and tracking system , some algorithms and methods are proposed and improved to solve those problems. The results of our experiments show that the methods and algorithms proposed in this thesis are adaptable and effective.