A Robust Video Stabilization Algorithm Based On Feature Matching

With the prevalence of camera device, video shot can be easily obtained. However, such video obtained in moving carrier has the shaking problem resulted from camera's shacking. Video shacking not only leads to bad video effect but also leads to false judgment or missing judgment,which are harmful to the next image object trailing. Thus,the require of stabilization of video is more and more important.The most common video stabilization framework including three modules: global motion estimation, motion filter and motion compensation. The object of global motion estimation module is to calculate the relative motion vector of adjacent frames' background, that is, the global motion vector. Global motion includes two parts, intentional motion and camera shaking. The motion resulting from cameraman's will is intentional motion, and the difference of global motion and intentional motion is camera shaking. For smooth visual effect, motion filter should be introduced to alleviate shaking. The function of motion compensation module is to translate, rotate and scale the video image according to the opposition of the shaking parameters so as to achieve smooth and stable visual effect. In this paper, a robust and intelligent video stabilization algorithm based on feature matching is put forward and realized. The main works in this paper are:(1)The traditional Shi–Tomasi method is improved by introducing an adaptive illumination independent model to realize robust feature tracking. Then feature matching errors and motion consistence principle are used to improve feature reliability. Experiment results show that the proposed method is robust to both moving object and illumination change.(2)The correspondence between matched features is used to shacking detection.(3)We developed an automatic video stabilization system based on our proposed techniques. The relative motion vector of adjacent frames'background is used to make sure whether shacking is exist. If shacking does not exist then motion compensation is not need. So, system efficiency reduce by motion compensation is avoid.