Research on Rotational Image Stabilization Algorithms for Electronic Image Stabilization System
|School||Harbin Institute of Technology|
|Course||Information and Communication Engineering|
|Keywords||electronic image stabilization affine model local motion estimation rotation motion estimation|
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 shaking. Video shaking 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.Some basal knowledge of electronic image stabilization are introduced. The most common video stabilization framework includes two modules: motion estimation and motion compensation. Traditional electronic image stabilization algorithm works when there is only translation between frames. Such as representative points matching algorithm, projection algorithm, features tracking algorithm and bit plane matching algorithm. When there is rotation between frames in video, it is bad to deal with the video. Now, there are mainly two kinds of method to deal with the rotational video. One is phase correlation in frequency field, the other is motion estimation algorithms based on affine model. And the latter method is emphasis of the dissertation.Based on the latter method three improvement schemes are put forward such as diamond search algorithm based on fully-distributed fixed block matching method, gray projection algorithm based on fully-distributed fixed block matching method and feature tracking algorithm based on fully-distributed fixed block matching method.By selecting one typical image, image rotational sequences data for simulation are generated, and the real image sequences data are obtained by digital video camera. The image rotational sequences data and the real image sequences data are processed by compiled electronic rotational algorithms’matlab program on which the discussed three rotational stabilization algorithms are realized. According to the definition of quality factor PSNR, the three algorithms’ accuracy and the image stability degree after compensating the translation and rotation vector of the three algorithms are evaluated. The three algorithms’ results are compared and the scope of application is given respectively.The simulation experiment results show the efficiency of the electronic rotational stabilization algorithms that the dissertation proposed. After processing the stability of the dynamic image is improved, the translation and rotation jitter of the video sequences are eliminated.