The Research Of Moving Target Detection In Dynamic Background

Moving target detection in dynamic background is one of important subjects in computer vision. It has broad application foreground in guided weapon, traffic monitoring and intelligent vehicle. The research of moving target detection under the condition of background moving caused by the camera motion has important theoretical value and practical significance. The algorithms of dynamic background compensation and optical flow are researched in the thesis.The methods of traditional background compensation were analyzed, in which there were problems of search slowly from a frame to a frame in gray projection algorithm. In order to solve these problems, the gray projection background compensation algorithm is proposed based on least-squares forecast. Binary search is applied to gray projection and linear prediction is used in background moving compensation. Thus, the efficiency of background compensation is improved.In the area of moving target detection, the algorithms of Horn-Schunck(HS) optical flow and Lucas-Kanade(LK) optical flow are researched. Aiming at the problem of inaccuracy and noise sensitivity in HS and LK, the improved LK-HS optical flow based on gradient threshold is proposed. LK optical flow idea and gradient threshold are applied into classical HS optical flow calculation. So, the optical flows of moving targets can easily be identified from the optical flows of background. Experiment results show that the integrity and accuracy of moving target detection can be improved.At last, an improved moving target detection method is proposed by combining gray projection background compensation algorithm based on least-squares forecast with gradient threshold LK-HS optical flow. Experiment result shows that the improved method can easily identified the moving target from background and effectively increase the accuracy and robustness of the moving target detection in dynamic background.