Research On Vehicle Detection Tracking And Classification Technology In Traffic Video

With the rapid development of society and economy, traffic becomes more and more important to the humanity. The Intelligent Transport Systems (ITS) is an important measure to improve the traffic management's level and quality, and is closely related with people's lives. Vehicle detection, tracking and classification technology provide a good way to collect and analyze the dynamic traffic information in the ITS and can be widely used in the areas of vehicle charge, road monitoring, large parking lot, road's efficiency improving and so onThis thesis does research on vehicle detection, tracking and classification technology based on the traffic video, and proposes corresponding algorithms and solutions. The main contributions in this thesis are as follows:(1) Through the analysis of existing moving detection methods and background model, this thesis proposes a background reconstruction algorithm based on frames difference and adaptive background updating algorithm based on multi-informational fusion.(2) To solve the problem that the ordinary object tracking method is difficult to get the accurate tracking result under the intricate conditions, such as scale modification, rotation, noise interference and so on, this thesis proposes a Mean Shift object tracking algorithm based on SIFT descriptor(SIFT-Mean shift).(3) We proposes a staged classification model according to the classification standard, select some vehicle features and demonstrate their distinguish ability and effectiveness, and then proposes a moving vehicle object reorganization method by their dynamic features.(4) This thesis designs the appropriate classification model according to the characteristic of different classification stage. In the first stage, we use the improved genetic algorithm to optimize the neural network's structure and weights to avoid the neural network falling into local optimal solution. In the second stage, we design a fuzzy neural network after getting the membership function which is optimized by genetic algorithm. The trained fuzzy neural network can be used to classify the vehicle's size. Moreover, this thesis also designs and performs several experiments on the proposed methods mentioned in the thesis. The experimental results show that these methods are feasible and effective.