Design And Software Implementation Of Electronic Image Stabilization Based On Motion Estimation

Electronic image stabilization is a kind of technology to stabilize jitter video sequences with digital image processing entirely. Compared with the shortages of traditional stabilization technology, such as enormous volume and high cost, electronic image stabilization has many advantages, such as good operation, high speed, compact size and reasonable price. It has been widely applied in military fields and civilian fields, and paid more and more attention to.This thesis expounds the research background and significance of electronic image stabilization, reviews the research status of domestic and aboard, and describes the basic principle, the motion model and primary procedure of electronic image stabilization system. This paper analyses several types of classic motion estimation algorithms--block matching algorithm, gray projection algorithm and representative point matching algorithm, including their principle, characteristic and application situation. Besides, this paper analyses theoretically two kinds of motion compensation and introduces several evaluation criteria for stabilization performance.According to no obvious gray change in image area influence on motion estimation, based on analyzing and making research on the sum of absolute difference between adjacent blocks in areas to be matched from the view of image feature, this paper designs a method of block classification. Experiment results show that image areas with obvious gray change can be selected by the proposed method. Compared with full search block matching algorithm, the method improves the computing speed by four to five times and the accuracy of global motion estimation effectively.According to the mismatch using gray projection algorithm under changing illumination intensity circumstance, gray projection algorithm based on difference between gray projections has been designed. Based on computing the values of row projection and column projection in searching blocks, the differences between adjacent row projection and column projection are utilized to search the best block. Experiment results show that the stabilized image can be matched with reference frame by the method, even though two frames are obtained under different illumination intensity circumstances.In this paper, we realize the electronic image stabilization algorithm combining block classification method and the improved gray projection algorithm under the environment of VS2010+OpenCV; we also make contrast experiment for four sets of different search blocks and five sets of different illumination intensity, and analyze the stabilization performance. Experiment results show that due to balancing the completeness of image areas, peak signal to noise ratio of current frame after stabilization with bigger search blocks is higher than that with smaller ones. And the improved gray projection method can make current frame and reference frame be basically in coincidence even under different illumination intensity circumstances.