Moving Object Attention Selection Using Optical Flow And Pulse Coupled Neural Network

At present, moving object detection acts a significant part in plenty of video surveillance systems. However, it is still challenging to obtain both the reliable accuracy as well as the fast processing speed. Therefore, a novel moving object detection model is addressed in this paper, employing Unit-linking PCNN, optical flow and topological property. Optical flow, which includes abundant motion information, has already been extensively studied for moving objects detection in the last decade. In this approach, Unit linking PCNN whole filter is used to represent the connectivity and integrity, which are significant topology properties, in moving object attention selection. Because of the smoothness ability of Unit-linking PCNN, the proposed approach can improve the quality of optical flow. False results can be eliminated to some extent and the detection accuracy improves. For video frames, we combine phase spectrums of topological property with color pairs by using quaternion Fourier transform to obtain saliency map. The experimental results on three databases reveal that the proposed approach reaches a higher detection accuracy and faster speed than FT(Frequency tuned salient region detection), GBVS(Graph-Based Visual Saliency) and PQFT(phase spectrum of quaternion Fourier transform).