Research On Intelligent Video Target Tracking Method Based On Human Visual Characteristics

In recent years,Intelligent Video Surveillance（IVS）has become the focus of attention in the field of artificial intelligence academia and engineering.With the rapid development of computer vision technology based on human visual characteristics,intelligent analysis technology and video surveillance are closely integrated.The traditional technology is to separate foreground objects from background by using image segmentation methods such as inter-frame difference method or background difference method for video capture frames,and to continuously track foreground objects in the surveillance area.In this paper,the detection and tracking methods of moving objects in static background video images are studied.Static background video image refers to a sequence of images with the same background when the camera is fixed.The first innovation of this paper is that in the stage of computer video acquisition and detection,the traditional method is to build a Gauss mixture background model for video signal,and use the background difference method to get the foreground binary image including shadow^[1-8].Aiming at the problem of low background separation of video frames,in order to get high-quality ontology image of the moving target by removing the shadow to the greatest extent,this paper proposes a local shadow detection algorithm combining texture features and YCbCr color features,which detects the shadow image,subtracts the shadow image from the foreground binary image.After filtering and contour filling,the foreground target with high quality is obtained.The second innovation of this paper is that in the target tracking stage,the traditional Camshift algorithm can not achieve accurate target tracking when confronted with background color similar interference and target is seriously occluded.Aiming at the interference of background color similarity,a Camshift algorithm is proposed,which combines two-dimensional H-S target color histogram of saturation component and hue component.The improved Camshift can effectively overcome the interference of background color similarity and improve the accuracy of target tracking.Aiming at the problem of serious occlusion of surveillance target,the improved Camshift and Kalman filter are combined,and the judgment mechanism of surveillance target occlusion is proposed.By analyzing the distance difference between the Camshift algorithm and the Kalman filter,and the ratio of the unmasked area to the initial area of the target,we can judge whether the target is seriously occluded.When the target is seriously occluded,the target motion position calculated by Camshift is very inaccurate.The Kalman filter's predicted estimate is used to replace the Camshift's calculated target motion position as the observation value.The parameters of Kalman filter are updated and the target position in the next frame is estimated.When the tracking target is reappeared in a large area,the target motion position calculated by Camshift is used as Kalman's observation value,and the parameters of Kalman filter are updated to estimate the target position in the next frame.Through algorithm simulation and scene test,the experimental results basically meet the expected goals.In order to optimize the application of intelligent video surveillance technology,innovative methods and practical ways are put forward.