Automatic Classification Of Video Objects Based On Support Vector Machine

Intelligent video surveillance system is one of new arising high-tech application fields. It can automatically analyse sequence image by the methods of computer vision and video analysis. The system can real-time detect, recognize, and track moving objects in a special environment. Furthermore, it can also analyse and judge the behavior of objects, hence to guide the action and make some decision. Intelligent video surveillance technology will improve the degree of automation of traditional video surveillance system. It has significance for economy, national defence and public security.Automatic video object classification is indispensable for intelligent video suiveillance. Categorizing detected video objects into semantic groups is crucial to analyze their activities correctly. This paper mainly discussed the design method of the structure of automatic video object classification system and the technology of classification feature selection and the technology of video object classification based on Suppor Vector Machine. The key problems below are solved in the process of finishing the study:(1) A general and extensible system framework for automatic video object classification is proposed. The system consists of training module and classification modele. The main ruction of training module is to generate a classifier by using proper machine learning method, based on plenty of video object samples. The classification modele categorizes the type of unknown classes' video objects.(2) A method to evaluate and select features for video object classification is presented. It determines which subset of the features should be chosen in video object classification task by applying the Adaboost method to compute a metric which indicates how well the features contribute to the classification performance.(3) Based on suppor vector machine method, the modeling of video object classification is realized. The particular approach to classify video objects by SVM is proposed.(4) Using the above methods, a video object classification scheme is implemented, which is able to categorize detected video objects into pre-defined groups of pedestrian, human group, vehicle or bicyclist. The examples sustain that the method of feather selection based on Adaboost and video object classification based on SVM proposed by this paper is rational and effective.