| With the development of computer technology, video surveillance systembecomes more popular. One of the most important parts of the surveillance system isto analyze the moving target in the scene, including detecting and tracking anomaloustargets.In fact, moving target detection and tracking has also been widely used inmany other areas such as industrial inspection, military guidance, artificialintelligence. Researchers have devoted themselves to this problem for decades andpresented a large number of moving target detection and tracking algorithms. Butthere are still many problems need to slove, for example the targets is occluded in thevideo. This paper tries to outbreak these problems. The main work includes:Research on moving target detection. Firstly, several moving object detectionalgorithm: inter-frame difference, background subtraction method, singleGaussian model method, ViBE method (background extraction method) areintroduece. Secondly, we analyze each algorithm theoretically and show relatedexperiments. Finally, based on the analysis and observation, we present acombination of single Gaussian model with ViBE algorithm. The sensitivity ofthe single Gaussian model can make up the ViBE algorithm’s undetected defectswhen detect the slowly moving objects. Experiment results show that thealgorithm can be implemented quickly and detect moving targets exactly.Research on moving target tracking. Current popular algorithms on targettracking includes Kalman filtering, MeanShift, CamShift and ABCShift. Kalmanfiltering algorithm is based on filtering theory, MeanShift algorithm is based onthe iterative search, CamShift algorithm is based on color features. If a movingtarget is occulded in the scene and there are many interferences are similar totargets in color, these algorithms will fail. Therefore, we presented a newalogrithm based on ABCShift and Kalman filtering. The experiments show thatthe new algorithm can track moving target effectively in complex scenes.Developing a moving target detection and tracking system. The above algorithmsare implemented by Visual Studio2008on Windows platform. In addition, somefuntions of OpenCV is also used in the system. The system is tested repeatedlyon popular videos and special videos. The test result shows that the system isrobust and accurate. |