The Detection Of Target In Moving Background And The Implementation On Embedded System

Moving target detection technology is one of the key technologies of computervision system and an important foundation of the target tracking and behaviorrecognition. At present, the target detection technology in the static scene hasapplied in military, transportation, industrial manufacturing and other fieldsmaturely. Compare to target detection in the stationary background; target detectionin the motive scene is more complex, so the target detection in a static scene cannotbe directly applied to target detection in the moving scene. How to detect themoving targets in the motive background accurately is the difficulty of targetdetection study. In this dissertation, we analyze the shortcomings of the currentdetection algorithm, propose the improved algorithm, and design a new set of targetdetection system.Firstly, we summarize the definition, the calculation methods and the principleof optical flow algorithm in detail, especially detailed analysis and introduce themotion compensation and the search algorithm of the block matching optical flow.And we study the block-matching optical flow method how to apply to the programof moving background.Then, we discuss the defects and deficiencies of block-matching optical flowmethod applied to the moving background, and propose an improved algorithm: weput forward an improved search method of block-matching algorithm which isadded the block-matching offset probability as an additional weight value. Thisimproved method can not only searches for the point near the center of the offsetwith a high probability, but also takes the large area of the offset distance intoaccount, so the overall search efficiency is greatly improved. For the compensatedmotion vector, we introduce an enhanced Ostu algorithm to automatically obtainthreshold, which used for binarization processing in vector to separate thebackground motion vector and the target motion vector. We apply the improvedmedian filtering to Canny edge detection algorithm, make it applicable to andintegrate it to the proposed algorithm, and then extract the target contour of motionregions. The simulation results show that the improved algorithm of this article canaccurately detect the moving object in a moving background. Finally, this algorithm is ported to the embedded platform. We analyze the TI’sDaVinci embedded platform firstly, describe the advantage of its dual-core parallelprocessing technology in the field of audio and video applications, and then designa set of video capture, processing (target detection), codec functions as the movingobject detection system and achieve the algorithm program to this target detectionsystem. Because the platform does not support floating-point algorithm, we convertfloating-point to the vertex in the transplant process. According to thecharacteristics of the hardware platform, we also optimize the program in theprogram implementation process. We apply the improved optical flow algorithm tosystem we designed. The experimental results shows that the proposed algorithmcan meet the requirements of the moving object detection in a moving background,and verify the feasibility and correctness of the method proposed in thisdissertation.