Research On Vehicle Classification In Video Surveillance System

Vehicle classification based on Video Surveillance Systems, as a branch of Intelligent Monitoring System, has already become one of the research hotspots. Vehicles classification which acts the basis of traffic management, transportation fares, traffic scheduling and statistical not only has important theoretical value, but also has broad application prospects. Vehicle classification method that this paper researched has two stages. The first step extract vehicle image from video sequence. In addition, the method which this article studied uses the feature which extracted from the vehicle image to classify the vehicle.The main work is as follows:The classification method this paper discussed extracted vehicle image from the video frame and did size normalization processing. In this thesis, it analyses image Hu invariant moments feature and moments feature extraction. At the same time, Hu moments fast algorithm has been introduced on vehicle classification. However, vehicle classification using Hu moments feature only able to distinguish the car and non-car. The classification results between truck and bus is unsatisfactory. At this point, texture features has been put forward and studied in this paper so as to improve the discrimination between truck and bus. Took the actual situation into account, Local Binary Pattern is used to extract texture features of vehicle images. Traditional LBP algorithm has been enhanced in dealing with the color image. LBP algorithm which improved has the ability to describe color characteristics. At the same time, according to polar coordinates mapped create the partition model LBP algorithm to solve the vehicle images with the same local texture, the same color, but the different global distribution. Then, two types classification methods Artificial Neural Networks and Support Vector Machine had been compared in vehicle classification. By comparing two classification methods Support Vector Machines had been proved to own superiority in this article. Soon afterwards, the strategy of SVM classifier extended to multi-class classification has been ameliorated.In the end,2400 testing sample images has been use to test the classification accuracy with the method which this paper studied. The Experimental results show that the method put forward has robustness in vehicle classification.