Research On Moving Object Detection Algorithm Based On Visual Surveillance

Thanks to the rapidly increasing number of surveillance cameras that leads to a strong demand for automatic processing methods for the surveillance data, visual surveillance has become an important research area in computer vision. Visual surveillance is a component of safety protection system; its main tasks include moving object detection, object classification, object tracking and object activity understanding. Moving object detection is an important step of processing video data in visual surveillance, so the performance of moving object detection algorithm directly affects the performance of visual surveillance. This work mainly focuses on the research of moving object detection algorithm.We analyzed the main moving object detection technologies, optical-flow algorithm, frame difference method and background subtraction method, and summarized their advantages and disadvantages based on experiments. Then we made further study on background subtraction method, analyzed common background models of this method, made experiments in different scenes and evaluated the results using evaluation system of background subtraction method. Through these analyses, we found that ViBe algorithm, self-organizing background subtraction algorithm and Gaussian mixture model had better comprehensive performance than other algorithms.Ghosts in detection results are caused by sudden movement of objects. They are regions of connected foreground points that do not correspond to any real object in the results. So they should be integrated into background as soon as possible. Aiming at the problem that ghost elimination of ViBe is too slow; we proposed an improved ViBe algorithm that combined with frame difference method. Improved algorithm uses frame difference technology to record the temporal changes of pixel values to distinguish ghost pixels. The experimental results showed that the improved algorithm eliminated ghosts faster than original algorithm, and had a higher accuracy. Aiming at the problem that fixed threshold of ViBe can not reflect the status of each pixel; we proposed an improved algorithm with adaptive threshold, according to the improved algorithm, each pixel has its own threshold. The experimental results showed that the improved algorithm had better performance.The shadows caused by the moving objects in a certain extent affect the accuracy of detection results. We studied the existing shadow detection technologies and analyzed their experimental results. Aiming at the problem that shadow detection technologies are not combined with background subtraction methods, we proposed an approach that was able to combine shadow detection technologies with ViBe algorithm. Experimental results showed that the proposed method could effectively eliminate the shadows and improved final detection results.