Studies On The Technology Of Vehicle Matching And Vehicle Tracking

Because of the rapid development of Intelligent Transportation System in recent years, vehicle matching and vehicle tracking which have utilized the knowledge and technology in computer image processing, pattern recognition, neutral network, artificial intelligence, genetic algorithms and the related fields have been used widely. In practical application, images are likely to change by perspective, light, scale, translation, rotation and affine, so it is very important to select the appropriate image feature and feature descriptor. The SIFT algorithm adopted by the present study can solve the problems of shading, rotation, light and affine which can cause changes in vehicle matching. The core issue of vehicle tracking is object matching. The present research realized several tracking method in different scenes based on the analysis and studies on SIFT algorithm, Kalman Filter, Mean Shift algorithm, and Gaussian background modeling algorithm. The author of the thesis has done the following work:1. Image feature extraction, description, moving target diction and tracking methods are introduced. The present commonly-used feature detection algorithm, Harris and Movers corner detection and SIFT algorithms are analyzed and compared.2. Through the studies on SIFT algorithm, improvement has been made on the high-complexity and slow computing speed of SIFT algorithm so that it can be better applied to vehicle matching. Experiments are carried out based on large amount of pictures, the results show that the improved algorithm can raise the matching speed as well as keep its stability, and can be successfully applied to vehicle entrance detection and vehicle matching identification demonstration system under the influences of partly shading, translation, rotation, and light.3. By analyzing the target tracking algorithm, for the lack of location update mechanism in Mean shift algorithm, the Kalman filter prediction algorithm is introduced to the Mean shift algorithm. Vehicle tracking are carried out under the combination of the two algorithms which can make the tracking results more stable.4. SIFT algorithm has the character of stability to the changes in image translation, rotation and scaling. Applying SIFT algorithm to the field of Vehicle tracking, together with Kalman filtering, can make vehicle tracking more stable.5. Multi-objective automatic vehicle tracking algorithm is a hot issue in computer vision field, the application of Gaussian background modeling can extract the background sequence and prospects of vehicle effectively, making multi-objective automatic vehicle tracking possible.