Video Vehicle Detection Under Camera Motion

Video vehicle detection is a kind of technology to extract the moving vehicle object in the video sequence.It is widely used in video monitoring and intelligent traffic detection system.Due to the characteristics of complexity,variability and so on,the technology is still in the initial stage,and it needs to be researched and improved continuously because of the vehicle vehicle detection technology,especially the video vehicle detection under the camera movement condition.In order to accurately detect the vehicle under the moving camera,this paper mainly do the following work:(1)Learn the theoretical basis involved in this article.This paper summarizes the domestic and foreign literatures related to the video image processing,especially the video object detection and so on.Choose and learn the relevant knowledge of the video vehicle detection under the condition of the camera movement,and focus on the research background and the results of the video camera detection in recent years.According to the video vehicle detection research status,combined with my own knowledge accumulation,and builds the construction of the algorithm framework.(2)Study and analyze the video vehicle detection method.According to the camera move or not,this paper introduces the method of video vehicle detection under static camera,and introduces the method of video vehicle detection under the condition of camera movement.(3)Design the algorithm of video vehicle detection under camera motion condition.The algorithm is divided into four parts.The first part introduces the algorithm of global motion estimation and compensation,which needs to be used in this paper.The basic algorithm is described.On this basis,by analyzing the advantages and disadvantages of different global motion estimation methods,six-parameter affine transformation model is proposed to affine transformation parameters,and then do the compensation after affine transformation of the image.The second part introduces the Gaussian difference algorithm and improves the differential step after motion compensation in order to obtain better experimental results.The third part introduces the nonparametric kernel density estimation,which is used to optimize the experiment.The fourth part describes the use of rectangular box on the detection of the vehicle target positioning.(4)Verify the algorithm of this paper experimentally.I use VS2010 software,Matlab software and use OpenCV library,use the video,which is captured by the moving camera as input.Write the algorithm of the experimental simulation program,and then do the experiment.The experiment is divided into two parts: the first part of the experiment used to verify the function of this algorithm;the second part of the experiment is a comparative test,used to verify the accuracy of this algorithm and high robustness.There are two points in the research and innovation:(1)In order to reduce the influence of the camera motion on the motion target detection in the video,improve the accuracy of the motion estimation.,when the affine motion estimation is carried out,we use different affine transform matrix for the frame before and after the target frame,the calculation is reduced,the affine transformation effect is improved.(2)Using the nonparametric kernel density estimation to optimize the proposed target to reduce the target hole problem and the noise sensitivity problem caused by each link.The shortcomings of the study are that the acquisition of the affine motion estimation parameters,the calculation amount is large,the time is long,the acquisition speed of the affine motion estimation parameters still needs to be improved,and the weather adaptability of the algorithm needs further research and improvement.