Research Of The Techniques Of Moving Object Identification And Tracking Based On Machine Learning

Intelligent video surveillance(IVS)technology is an emerging research orientation in recent years,It plays an important role in traffic video monitoring,aerospace,computer vision,medical image analysis and so on.Many researchers has made huge progress in this research,but there are still some problems to be solved in the practical application.This paper mainly research on moving object identification and tracking technology with the purpose of achieving more practical application significance and economic value of the IVS system.The main goal is to realize anti-interference and anti-occluding specific moving object auto-identification and real-time tracking unity under the complex background.Firstly it detects moving objects from the video according to the corresponding algorithm,then it makes the identification of the object based on machine learning,and finally it does some real-time tracking by concurrent and collaborative work strategy.The main three aspects of algorithm research are as follows:Detection of moving object: by the classic detection methods,it proposes a algorithm which merges codebook algorithm and three-frame difference.The method uses codebook to learn the background and the three-frame difference is employed to extract initial foreground object,then we adopt the processes based on Log edge detection and component filling.It associates the foreground object obtained by codebook with the object,which is obtained by improved three-frame difference to perform logic “AND” operation,and gets the final moving object.It can effectively obtain the entire moving object and enhance detection’s accuracy and reliability.Identification of moving object: It adopts vehicle and pedestrians classified identification method based on HOG feature and support vector machine(SVM).In this paper,it collects two kinds of samples,vehicle and pedestrians.And then by means of the HOG feature extraction,it learned the samples to find support vector and establish the optimal separating hyperplane.At the last,we can classify the pedestrians or vehicles of video by using the support vector machine classifier.Tracking of moving object: Discussing the basic principles of several traditional tracking methods,we propose a strategy of detection and tracking parallel and collaborative working.Detection and tracking can be divided into two threads simultaneously,and the data is transmitted in the fixed node.The object information which obtained by regularly detection thread is transmitted to tracking thread.This strategy is used to adjust object in case of lost and obscured,and finally we can achieve automatically capture and real-time tracking of moving object unity.