Research On Algorithms Of Moving Object Detection And Their Application

The modern society is becoming more and more complex. The security situation people face is becoming more and more serious. In this case, the intelligent video surveillance technology has been widely used. As the key technology on intelligent video surveillance, algorithms on moving object detection have gotten more and more concern from researchers and become a hot research topic of computer vision technology gradually.First of all, this thesis introduces the research background and current situation in domestic and abroad of the topic. Then, some classic algorithms on moving object detection have been researched. The advantages and disadvantages of these algorithms have been analyzed. This thesis analyzes some widely used method on moving object detection such as Inter-frame difference method, optical flow method as well as background difference method detailedly. Some staple technology on moving object detection such as image processing, shadow removal, morphological processing together with contour extraction also been researched in general.This thesis focuses on the research of the algorithm of moving object detection based on single Gaussian model. In this Thesis, the source of the single Gaussian model and the principle of the traditional algorithm of moving object detection based on single Gaussian model are introduced. The thesis points out the problem called "trailing" in the traditional algorithm of moving object detection based on single Gaussian model and analyses the root cause of the problem. Then the improved algorithm of moving object detection invented by koller is talked about. The algorithm invented by koller has solved the problem caused by the traditional algorithm of moving object detection. But it produces a new problem called "ghost". Based on the deep research of the cause of the problem called "ghost", the thesis invents a dynamic strategy to update the Gaussian model. With the new strategy, the problem called "ghost" has been solved well. A new proposal about how to select the turnover rate of the single Gaussian model is proposed in this thesis. In this strategy, the turnover rate for the mean and standard deviation of the single Gaussian model is different from each other. With this strategy, the astringency and stability of the single Gaussian model are greatly improved. In shadow detection, the thesis adopts a new algorithm to eliminate the shadow. In this algorithm, both the change information of the chroma and the first-order gradient information are considered synthetically. Some experiments prove that this algorithm can achieve good results. In the moving object contour extraction, the thesis carry out morphology process to the binary image containing the moving object contour firstly, and then the interference of smaller items removed. After these procedures, the moving object contours are extracted accurately. Centroid and the mean of the distances from each pixel to the centroid are calculated to get the bounding rectangle of the moving object contour. Then the location of the moving object can be identified accurately. Experiments show that the algorithm of moving object detection based on improved single Gaussian model can achieve good results in the indoor environment and outdoor environment with simple background.Based on the algorithm of moving object detection the thesis proposed, the thesis design a system of intelligent video surveillance for energy-saving which is a new direction for the application of the intelligent video surveillance technology.