Researches On Moving Target Detection Technology Under The Condition Of Camera Scanning

In recent years,the technology of moving target detection has been widely used in the field of computer vision,such as intelligent video surveillance,traffic management,and military action,etc.The traditional target detection algorithm can accurately extract moving targets in the condition that the camera is fixed.However,there are few researches on moving target detection under the condition of camera scanning,in addition,the real-time capability and reliability of the algorithm need to be further improved.In general,moving target detection under this condition is using affine model or projection model combined with parameter estimation method to compensate the moving background,after staticize the dynamic background,the moving target detection can be realized by the detection technology in static platform.Nevertheless,after analyzing and visualizing the motion model of the video image sequence,we found that the motion model is not any form of affine model or projection model.Due to the problem of non-uniform distortion caused by planar imaging mechanism,using the affine model or projection model to approximate the motion model of the camera will cause significant error.In addition,video image processing requires high real-time performance,and there is a limitation of the camera system hardware,so the algorithm also needs to have high real-time performance.Therefore,how to realize the real time and robust detection of moving target under the condition of camera scanning still faces great challenges.Due to the limitations of existing motion compensation algorithms,this paper presents a method of moving background compensation based on camera scanning model.Firstly,we establish a camera equation under the condition of camera scanning,and linearize the equation.Then,we use the point line duality of the image space and the parameter space,convert the linear detection problem in the image space to the parameter space by using Hough transform,so that we can estimate the equation's parameters quickly and robustly.This method can solve the problem of the background motion and the non-uniform image distortion at the same time.In the experiments,we verified the feasibility,robustness and accuracy of the proposed method.For the problem that the target is not complete when we use frame difference to extract targets,this paper proposes a method of extracting moving targets by using background subtraction based on cylindrical background model.It combines the camera motion model and the cylindrical projection model to establish a panoramic cylinder background model,which can accurately describe the scene and improve the quality of moving target detection.In the experiments,we verified the effectiveness of the proposed method.