Research On The Algorithms For Content-Based Vehicle Video Retrieval System

Content-Based Traffic Video Retrieval System is a human-effort-spared technique by which the vehicle features, such as the traffic volume, speed, color, shape of vehicles, etc., are to be extracted by processing and analyzing the traffic video captured by cameras. These traffic features are extracted and stored in a database together with the accompanied video, and then people can search for the information of a vehicle by specifying features on-line or off-line. Not only could this technique provide the first-hand statistical and essential data for metropolitan traffic planning scheme, but also could provide an efficient and convenient tool for a transportation section in investigating vehicles that break traffic regulations.In the thesis, a number of novel algorithms for Content-Based Vehicle Video Retrieval System are presented based on our researches and experiments. Our work contributes to the following aspects:(1) Detection of moving vehicles. We propose a dynamical method of renovating the model of background, and provide a better solution to the problem of video shaking unsteadily captured by camera. We also work out an algorithm to detect the vehicle shadows based on Gauss-probability, and a method of detecting vehicles based on the outline of shadow. The experiments showed that using of these algorithms is comprehensive and effective.(2) Tracking of moving vehicles. We match the moving vehicles by combining time and spatial characteristics, and classify the problems into the following three cases: well matched, shadowed, and joined in of a new vehicle. For each case, we give a strategy to handle it. The experiments showed that using the above strategies we can follow multi-vehicles effectively and steadily in real time.(3) Extracting and retrieving the features of vehicles. We work out a method base on extracting the key frames of the traffic video for moving vehicles. Using the method and combining with some reasonable pre-assumed information about the traffic flow, we can acquire the features of vehicles such as color, shape, size, speed, direction, the time of getting in and out, etc. Our experiments showed that we can get comparative exact information of the vehicles we need.