Study On Target Detection Of Non-stationary Camera

Motion estimation is a primary problem of non-stationary camera target detection.Motion estimation and motion compensation can reduce the effects of camera global motion,which can significantly improve the target detection of non-stationary camera. Therefore,research on efficient and robust motion estimation algorithm has already been an importanttopic in the field of computer vision. Among current motion estimation algorithm,block-matching algorithm(BMA) has the advantages of simple principle and easy toimplement. But BMA also has the problem of complicated calculation, so it is a hot topic toimprove the speed of BMA. Traditional fast algorithms such as Three-step Search(TSS),New Three-step Search(NTSS), and Diamond Search(DS) have been proposed to improvesearch speed, but they have an inherent shortcoming of being liable to get trapped in localoptima, which can degrade the quality of motion estimation to some extent.On the basis of above-mentioned work, an improved adaptive motion estimation isproposed which has fast speed. The new ARPS based on motion pattern classification ispresented. The new algorithm firstly detects the motion pattern of current frame, thenpredicts the initial search point based on the spatial correlation in the motion vector field.And stopping criteria is introduced to improve the search speed. Finally experimentalresults show, the algorithm in this paper can achieve the desired goals in the searchingaccuracy and searching efficiency. Comparing to the traditional fast block matchingalgorithm, the proposed algorithm has batter performance in various types of video thanTSS, NTSS and DS, especially for the video sequences which have clear motion pattern.