Fast and reliable detection of moving objects isone of the important requirements for many computervision and video analysis applications. Mean shift basednon-parametric background modeling supports moresensitive and robust detection in dynamic outdoor scenes.However it is prohibitive to real-time applications such asvideo surveillance. This paper aims to deal with thelimitation of high computational complexity. Firstly, coarseto fine methods are proposed to avoid raster scanning entireimage. Foreground pixels are detected in coarse level toroughly locate the foreground objects in the image, and thenfine detection is performed on the corresponding blocksgradually. Secondly， fast mean shift approach is presentedaccording to temporal dependencies. Mean shift iterationsare performed starting from incoming data and the modesobtained last time. The experimental results show that theproposed algorithm is effective and efficient in dynamicenvironment. The proposed algorithm has been appliedto move objects detection in our real-time marinevideo surveillance system.
展开▼
