Research On Moving Object Detection And Classification Algorithm In Intelligent Video Surveillance

With the rapid development of multimedia technology,the video surveillance system plays a more and more important role in public safety.However,the traditional video surveillance system has greatly wasted manpower and material resources and completely failed to meet the development needs of the times,the video surveillance urgently needs to develop in the direction of intelligence.The moving object detection and classification has always been the key issue in the intelligent video surveillance.However,due to the complexity of the monitoring environment and the limitations of the object features,there are still many difficulties in current moving object detection and classification algorithms.Therefore,the corresponding research on moving object detection and classification algorithm in intelligent video surveillance is carried out in this paper.1.According to the Gaussian mixture background model and frame difference method,this paper proposes a new moving object detection method based on GMM(Gaussians Mixture Model)and frame difference.Firstly,we proposes a new image combination algorithm to combine GMM and frame difference,which solves the problems of noise and the inner cavity of object caused by the traditional fusion method based on GMM and frame difference.Secondly,we add image repair technology to the proposed moving object detection method,which compensated the moving object in space,and obtain a better target shape.Finally,this method uses morphological knowledge to process the moving object and obtains accurate moving object.Experimental results show that the proposed moving object detection method can effectively solve the problems such as incomplete detection,inner cavity and noise,and plays a crucial role in the subsequent moving object classification.2.In order to solve the problem of low recognition rate of single feature classification,this paper proposes a multi-feature fusion method for moving object classification.The multi-feature includes the static features and motion features.Static features include the aspect ratio and edges of the moving object and block-based HOG(Histogram of Oriented Gradient)features.The block-based HOG not only considers the direction of the gradient,but also considers the geometric position of the gradient,which makes the extracted target shape information more accurate.The motion feature is spatial entropy of optical flow,which is based on the direction and the intensity of optical flow,and its dimension is small.The feature takes the local motion direction and velocity into account,so it is a very good representation of the local dynamic characteristics of the object.Experimental results show that the multi-feature fusion method proposed in this paper has higher classification accuracy than other feature methods.3.In order to solve the multi-objective classification problem,this paper constructs a binary tree Support Vector Machine based on statistical learning as a classifier for moving objects classification.Due to the large feature vector dimension and large training data set,this paper uses the Batch incremental Support Vector Machine instead of traditional SVM(Support Vector Machine)to train the classifier,it can improve the classifier's learning efficiency,and solve the competitive classification problem of the traditional SVM.