Research On The Key Technologies Of Video-based Intelligent Transportation System

In recent years, with the rapid increase of vehicle possession, Intelligent Transportation Systems (ITS) have been of great importance for enforcing traffic management policies and investigated by many researchers. Because traffic video information has been collected and used as an important part of traffic surveillance and management, major importance has been attached to video surveillance technology which is based on video image processing, analyzing, and understanding, to improve intelligence of traffic surveillance and management. As a subject of general interest in area of image engineering and computer vision, video surveillance technique is different from traditional surveillance technique in that it is highly intelligent; therefore, research on this technique and its applications in video-based ITS are of great practical significance.Moving vehicle segmentation is an important and key part in video-based ITS. We proposed a new, simple and fast method to detect moving vehicle from video based on flood protection algorithm. Firstly, dams were built to protect vehicle regions and gray approximation was estimated from difference image. Secondly, all pixels belonging to background and shadow were removed based on estimated shadow grey value. And finally, we got moving vehicle images. Experiment results confirm that this proposed method is simple and fast. It also has good stability and validity.This dissertation proposed a novel vehicle tracking method based on corner features and Mean-shift algorithm to deal with the considerable change of tracking object in two adjacent frames. At first, seed windows of object model, which have the same size of kernel window, are built depending on corner features. Then, Mean-shift algorithm is carried out sequentially after initial locations of seed window's centroids are adjusted up to moving orientation characteristic of vehicles on the highway. According to tracking results, the final position of object is estimated based on a judgment rule. Applied to track vehicles that vary in size, the proposed method has brought out remarkable results. This method has three significant advantages. Firstly, it is efficient to track the vehicle targets which change remarkably in two adjacent frames by using corner features. In addition, it has high convergence velocity due to the adjustment of the centroid position. Finally, the implementation of this method is simpler and faster as tracking the large vehicle object because only a part of the object is tracked instead of the whole one. This method, however, is only suitable for the objects that have marked geometry features.(3) It is an important and key part for moving vehicle extraction in video-based ITS. In order to get a faster algorithm and good discrimination ability of vehicle classification, by analyzing the properties of the real fart of Gabor filter, first we simplified the calculation of two-dimensional convolution and used recursive implementation. Then we used segmented sampling to get Gabor features for classification on the basis of the edge features in vehicles. Experiment results haves confirmed that this proposed method is efficient and reliable, and its computation cost can also satisfy the real-time vehicle recognition system.At last, this dissertation points out some problems that need more research and development direction in the future.