Detection,Classification And Tracking Of Moving Objects In Video Surveillance

As the times progress and society advances,people pay more and more attention to the issue of social security.Video monitoring plays an important role on social security.With the continues development and update of the computer technology and the hardware technology that provides a full hardware platform and technical support of the intelligent development of the video surveillance technology.Detection and analysis of moving objects in video is one of the key technologies of intelligent video technology.In real life,most of the meaningful information in the video is included in the moving target.Intelligent video surveillance is the hope that through the computer to complete the processing of motion information.In this paper,the three parts of the intelligent video system were studied,such as the detection of moving targets technology,moving target tracking technology,a moving target classification and identification technology.The main contents of the research are as follows:1.In order to overcome the shortcomings that the parameters of the adaptive Gaussian mixture model,which is used for the background modeling is updated slowly,a new parameters learning mechanism is proposed in this paper.The learning rate of learning mechanism varies with the change of weight.Experiments showed that the learning mechanism can make the mean and variance learning faster.The detected target contains two parts,which are object and shadow.According to the characteristics of the shadow in the HSV space,a shadow detection method is proposed based on the improved adaptive Gaussian mixture model.The shadow detection method is controlling based on the component of Saturation and Value to implement the shadow detection.Experiments showed that the shadow of moving objects can be eliminated effectively,and the time consumption of the testing model is relatively small.2.In order to accurately extract the target features,a weighted“average” is proposed in this paper.Use the "average" to replace the classes sample mean of the kernel Fisher discriminant analysis for target classification.The "mean" is a weighted average of each image of each class.Firstly,with the continuous input image,the image weight is decreasing.Then the last three image weight are increasing.There are two experiments to verify the superiority of the algorithm.The first experiment is automatically classify the moving objects in video,and the second experiment is face recognition.Experiments show that this algorithm can quickly and accurately identify the target.3.In order to overcome the shortcomings of the traditional tracking methods need to adjust the threshold according to different background,a tracking method is put forward,which is combined adaptive threshold with kalman filter.Firstly,get each image column average after sorting,then take the 0.35 times of middle value as the threshold.The experimental results show that the method can track the target well.