Computational Methods Of Video Salient Object Detection Based On Biological Vision Mechanism

Research work in the field of computer vision has reached a climax with the rapid development of computer science and artificial intelligence.Compared with the excellent processing mechanism of the human visual system for external information,the processing capability of computer is still far from enough.Therefore,through continuous experimental research on human and animal vision systems,experts in the field of computer vision have proposed a large number of computer models in recent decades.These visual models are widely used in real life such as video surveillance and intelligent transportation,and most of them have achieved relatively good performance.At the same time,with the rapid development of intelligent monitoring,intelligent robots and drone navigation,target-based analysis and research work in images and videos have great value.This paper analyzes and researches the calculation model of biological vision attention mechanism,and proposes a video salient target detection algorithm based on biological vision mechanism.The main work of this article is as follows.First,with the deepening of the research on the visual mechanism of flying insects in recent years,scientists have found that flying insects do not use binocular stereo vision to estimate the depth of the scene for navigation because of the special compound eye structure,but use optical flow to control their various Flight behavior makes computer vision algorithms based on optical flow that imitate the visual mechanism of flying insects a hot topic of research.Therefore,in Chapter three of this paper,we propose an improved optical flow algorithm to calculate moving targets in video sequences.Firstly,we propose a definition of spatio-temporal gradient to guide the contrast calculation.It can assign high saliency values around the foreground target,while avoiding the difficult to overcome hole effect.Secondly,based on the comparison of the amplitude values of the motion gradient,we adaptively fuse the time gradient and spatial gradient information.Finally,we calculate the average optical flow histogram at the superpixel level,and extract the moving target by comparing and calculating the motion information between the superpixel blocks.Our algorithm can completely extract the moving target from the optical flow color map.In the case where the optical flow cannot separate the moving target and the background(the continuous frame hasno motion or the motion is too large to meet the basic equation of optical flow),we can also accurately extract the moving target according to the motion comparison between the superpixel blocks in the frame and the frame.Second,the research of biological vision attention mechanism is a hot topic in computer vision tasks.After analyzing and researching two theoretical models of visual attention mechanism in Chapter 4 of this article,inspired by the guided search theory,a video salient object detection model based on a two-pass framework is proposed,which simulates the top-down and bottom-up visual attention process.The model uses spatio-temporal contrast to guide the search for salient targets.First,along the non-selective path,the intra-frame and inter-frame mapping of color contrast and motion contrast,combined with the saliency clues of the previous frame,are used as a priori information on the spatial location of the target.At the same time,low-level features such as brightness,color,and motion are extracted in the selective path to achieve accurate search for targets.Finally,the improved Bayesian inference model is used to further obtain the optimal results.Experimental results show that,compared with the typical contour-guided visual search method,this algorithm improves the performance of salient object detection in video.At the same time,compared with seven classic video salient object detection models,our algorithm achieves relatively good performance.