Moving Target Detection Algorithm Research Based On Video Image Sequences

It is an important research direction for moving target detection andtracking based on video sequences within the field of computer vision. In this paper,based on study and research for previous theoretical, through carefulanalysis,absorption, and thus simulations of several existing classical theory areachieved. Emphasis on linking theory with practice in learning, Kalman backgroundestimation algorithm is successfully applied in the actual project, and theoreticalunderstanding deepened based on its applications and thus an improved Kalmanbackground estimation method is presented. In this paper, the work carried out can besummarized as follows:The first part of the paper analyzes the main de-noising algorithm, based on thesystem reseach for the image preprocessing, in accordance with the general imageprocessing methods and procedures. On the fusion of theory and practice, recognizingthe importance of difference in the pretreatment, doing difference between the sourceimage and the image after large scale filter is found to be well de-noising. Thistreatment method is used in cases, while the use of the difference in some vague ideasfeature extraction can obtain excellent results. The second of the paper describesseveral classical algorithms, and achieves the relevant theoretical formulation resultsby simulation experiment, understands the theory from practice and has a deepunderstanding the idea of algorithms. Then build the visual system, whose nature isapplication, through the actual project, and asked to select the appropriate Kalmanestimation algorithm combination of the actual practice and achieves the desiredresult of theoretical verification. In this process, improved Kalman backgroundestimation algorithm based on entropy is proposed, after in-depth understanding forthe Kalman background estimation algorithm. Through entropy characteristics of theimage, with the integration of the Kalman background estimation using entropythreshold to update the Kalman gain coefficient, and to determine the foreground andbackground and to achieve effective tracking of moving objects. Combines thecharacteristics of entropy, on the basis of the original algorithm a more stable result isachieved, which removes part of the shadow and background interference andimproves the robustness of the original algorithm.