Study Of Video Segmentation Algorithm For Moving Fish Based On Visual Saliency Driven

Video analysis technology has become an important means to promote aquaculture information and intelligence, video segmentation is the key which is performed to track fish, behavior monitoring to identify, and abnormal behavior early warning. Aiming at moving fish video segmentation under indoor aquaculture scene, the visual saliency driven moving fish segmentation algorithm is studied In this paper.First, a large number of moving fish video data is collected and recorded, and the characteristics of the underlying image brightness, color, texture and edge are analyzed to study the preprocessing algorithm for fish aquaculture scene surveillance video. On this basis, the basic theory and commonly used video segmentation algorithms of moving targets are studied. The facing challenges of moving object segmentation are described, which is the foundation for the study of subsequent video saliency driven moving fish video segmentation.Next, contrast and analyze of the classic Itti visual saliency computational model and graph-based visual saliency model. Focusing on the issues which are faced on the graph-based visual saliency model is applied to moving fish segmentation in aquaculture,an improved algorithm is presented.the improved algorithm is a fusion of multi-feature motion feature automatically weighted for panoramic video sequence to extract having significant interest in the dynamic characteristics of the area. Experimental results show that the improved saliency model can help to extract the regions which included moving significant on.Finally, further analysis of the actual problem which is faced on the visual saliency computational model used in farming scene to monitor. To suppress noise interference which are reflective surface clutter, circulating water and other non-moving fish, the bottom-up saliency calculation and task-oriented top-down prior information combined, based on support vector histogram of oriented gradients machine model, the extracted significant area of the underlying sub-graph are for targets verified. remove noise, and ultimately accurate segmentation of the moving fish in the video. The rationality and validity of proposed algorithm is verified by the experiments to segment moving fish targets in aquaculture monitoring video.