WMD-Based Saliency Detection Of Moving Object Under Free-moving Camera

The human visual system has a strong ability to locate and search objects of interest in the scene.This mechanism of how to obtain important information from massive multimedia data is called the visual saliency mechanism of computer tasks.For video saliency detection,the target is to identify the most significant object from the video sequence.Because of the saliency,the accurate detection of the saliency of moving objects in free-motion cameras with complex motion and complex scenes is a difficult problem in the field of computer vision.In the video stream of free-motion camera,the relative displacement of the background between frames and the moving target will lead to the change of local descriptor information in the moving target area,which will lead to weak matching.This paper use this local weak matching descriptor to represent the saliency of moving objects.This paper proposed weak matching descriptor(WMD)to describe the saliency of moving objects,and realized the saliency detection of moving objects under free moving cameras.This paper studies the above problems,and the main work is as follows:(1)Analyzed the generation principle of WMD,proposed the definition of WMD,described the extraction method and flow diagram of WMD,and preliminarily extracted the moving target area in the video according to the generated WMD.(2)A variety of different feature descriptor extraction algorithms are used to obtain WMD extracted in different algorithms.A joint decision making method is proposed to obtain the final feature description subset cooperation as the rough extraction result of the motion information.(3)according to the extracted WMD,the concave point detection algorithm based on the convex hull was used to outline the general contour of the moving target,and the extracted contour constituted a closed interval.The number of feature descriptors was increased again in the interval to enrich the internal saliency information,the superposition information of gaussian beam spot was used to represent the relative intensity of the saliency of moving target,and the closed operation of morphology was used to fill the void information of the target.(4)SLIC algorithm was used to fine segment the obtained motion information,and then the edge intensity information was extracted based on the global target saliency detection algorithm.The edge intensity information and the super pixel block size are fused to generate the global saliency graph with moving object priors.(5)The weight matrix with SLIC super pixel block as unit node and intensity value difference as edge is constructed,the weight matrix of target significance is updated by Dijkstra shortest path algorithm,and the shortest path from each node to the outermost node is obtained.The adaptive threshold is set to finally refine the significance of the moving target under the free moving camera.According to the above method,the moving object saliency detection is finally realized under the free moving camera.A large number of experimental results show that this method can accurately detect the saliency of moving objects and has good robustness and effectiveness.