Research On Moving Target Detection Method Based On Visual Saliency

This paper proposes a spatiotemporal saliency method to detect moving object for the videos captured by a moving camera.The initial probability distributions of the foreground and background was estimated by performing the manifold ranking algorithm with this boundary prior for each frame,the Spatial saliency map of the foreground object is generated.The optical flow was estimated forward and backward for each frame,the temporal saliency map is generated.Finally,after the spatiotemporal saliency map was computed,the segmentation results was refined at the superpixel-level into those at the pixel-level by employing the Markov random field optimization.Experimental results on extensive datasets demonstrate that the proposed algorithm outperforms the state-of-the-art techniques significantly.The paper has some main work:(1)The basic concept,classification,application background,significance of moving object detection is introduced,the research status of moving object detection technology are analyzed.(2)The human visual attention mechanism is introduced,and several existing visual saliency detection algorithms are analyzed.(3)Many challenges and difficulties were presented in details.The research contents and methods of moving object detection under static background and dynamic background are analyzed.(4)In view of the complexity of moving object detection in dynamic background,optical flow was estimated,and the manifold ranking algorithm was used to detect moving objects in dynamic scenes.Experimental results on extensive datasets demonstrate that the algorithm can solve such problem as moving cameras,dynamic background,illumination change,and image noise,the moving object was salient in the frame.The experimental results demonstrate the algorithm can simultaneously handle the video of static and dynamic background,and the effect of the algorithm is better than the comparison algorithm.