Research Of Moving Object Betection Algorithms And Tracking Technology

Ivs(Intelligent Visual Surveillance) is new research orientation in the filed of computer vision. It is based on tradition Visual Surveillance. Ivs make full use of computer vision to analysis and understand the surveillance information. According the situation to control the monitoring system effectively, the monitoring system have intelligence. It can help people to finish some work. Intelligent Visual Surveillance is a complex system. It is hierarchical, In the low layer there is moving object detection, In the middle layer there is moving object tracking and classification, In the high layer there is moving object behavior understanding and analysis. Intelligent system have a wide range of application.No matter in the military areas or in the civilian areas play a significant role. Take traffic management battlefield surveillance missile tracking for example.In the system, moving object detection and tracking is the basis to anther layer. There two technologies are the key of this system, It is play important role to this system. This paper is analysis the video streaming of moving object detection and tracking,under the situation of the camera in static. In fact, the video stream is consist of image sequence. the task of us is to find the moving object in the image, and to classified with the background, extract prospects which have the moving object. Then we can tracking he moving object in the video image and get the moving object's motion parameters.In this paper, we first learn exist detection and tracking algorithm image process particle filter and GMM, Then we research a large number of the relevant literature. Based on this. we introduce and analysis these algorithms. Experiments showed that good effects had been achieved.First of all:we researched and analysis these moving object detection algorithms which used in exist system. From this, we analysis and summary these algorithm's advantages and disadvantages and application conditions. We mainly research these algorithms which the camera is in static. In this situation we mainly research and analysis the GMM which is belong to background subtraction. Then we improve the GMM based on it's disadvantage. Traditional GMM model is very sensitive to light, when the light change, the model will fail. For this problem, we propose a new failure detection method. and we use three frame difference instead of GMM. For the slow moving objects into the background and the stationary began to move. I proposed a new parameter update method in this paper. The traditional update method is that all pixels with uniform parameters update method. In this paper. we separate foreground regions and background regions. different regions can use different parameter update method. The parameters of the background update method not change. When the moving objects update its parameters, we can adjust the weight by a factor to make it not change big when update the parameters. So it can prevent the moving object the moving object change into the background. We propose a new method to process the problem, which the moving object change to stable. And we compare and analysis several commonly used moving object detection algorithms and find the best situation which is fit for the algorithm. Last we do experiment to these algorithms. The result showed that the improved algorithm is effective.Secondly:According to the feature of the foreground detection experimental data. We proposed an image post-processing denoising method. Then we extract the moving object movement features, then feedback these features to the prospect help to improve the moving object detection algorithm. In practical applications. Many factors influence foreground detection and make the image have noises. Take camera flutter because of wind. the camera deviation and the algorithm itself disadvantages for example, Because of these factors the video images collected have many noises. So eliminate noises is a critical step, In this paper we adopt conditional dilation operation. This algorithm not only can eliminate background noise also can keep moving object information. Then we can use opening operation and close operation to process the image. The experiment showed that the algorithm is effective.Last:In this paper we analysis and research some exit tracking algorithms. Based on this. We use and improve particle filter algorithm tracking the moving object. Real time is a key feature to the Ivs system. But now, the image resolution is improving gradually. So in the foreground detection we must spend more time to process these image pixels. This have a large influence on the real-time system. Based on this fact, we can feedback the moving object's position information (get from the tracking phase) to the foreground detection. Through this we can decrease the time which spend on the moving object phase. And the result is that we can improve the real-time of the system. The feedback improved based on the particle filter algorithm. Through the filter information we can get the moving object's focus. In chapter three we have extract the moving object's feature information. Then through these information we can get the area which the moving object in. Because of the background is stable. We can not update each frame, we can interval of a few frames to update the background, between this time,we can only update the active area which moving object in, not the hole image. Then the foreground detection process time will be decreased. The system will be more effective.In sum. We improve the algorithm base on the exit algorithm. And we compare and analysis these exit detection algorithms. According to the experimental data we adopt the conditional dilation operation the image process and extract the feature of the moving object which we needed. Last we decided to adopt GMM model and particle filter to construct a fast detection and tracking system. Whit the development of information technology. The Ivs will be mature gradually. I hope this paper will be helpful for this.