中国科学院自动化研究所机构知识库

Motion detection plays a crucial role in most video based applications. A particular background subtraction technique called ViBe (Visual Background Extractor) is commonly used to obtain foreground objects from the background due to its high detection rate and low computational complexity. However, the performance is not very satisfying. Therefore, this paper presents an improved ViBe algorithm to increase the accuracy and robustness of motion detection. Specifically, a foreground feature map is created by optimizing the result of ViBe algorithm. Then the edge detection of the original video frames is achieved after pre-sharpening using improved Sobel operator and Otsu algorithm. Finally, by feature fusion (of the foreground and background feature maps) and contour filling, the motion detection results can be obtained. The experiments demonstrate the improvements of the proposed modifications at a limited additional cost.