Research On Techniques Of Video Stabilization

With the development of technology and the popularity of electronic devices, we can use the video capture device to record our life more and more easily. However, due to the lack of professional shooting experience and stable equipment, the video is usually accompanied with jitter, which makes our vision inconvenient. Also, because of the increasing online video resources and the higher requirements for the video quality, video stabilization technology becomes more and more importance in our daily life.The purpose of video stabilization is to remove jitter information from video in order to improve the human visual feelings. Currently video image stabilization technology which takes the role of pretreatment technology in other video processing technology, has been widely used in the civilian and military applications.Traditional stabilization method consists of three steps: motion estimation, motion filtering and motion compensation. All along, video stabilization algorithms have difficulties in predicting the motion information precisely, especially in the presence of complex scenes, violent shake and respective moving motion of the foreground and background scenes. Thus how to capture the shooting intention is the premise of video stabilization. The main task of motion filtering is to remove overall jitter noise. In the existing methods, the majority algorithms could only smooth the global movement instead of filtering out jitter noise thoroughly. Therefore, the difficulty of motion filtering is to estimate the real-time noise and then filter it out.This paper firstly studied literature at home and abroad, and propose solutions for sloving current problems. Related exploratory work includes:1. Propose the concept of motion vector selection based motion consistency estimation and build a global motion estimation module based on motion vectors selection. Through the introduction of image feature points, it could avoid the limitations of single, inaccurate motion estimation. Also, we could capture the photographer's intent through motion vectors selection based motion consistency estimation. Further more, we could select feature points on this basis which would reflect shooting intention. Smooth and compensat the movement of these feature points, we can remove video jitter noise in the meanwhile maintaining the shooting intention of the camera;2. Improve Kalman filter in order to make it adjusting its parameters adaptively according to the noise in the system all by itself. In the proccessing procedure of image stabilization, the Kalman filter assumes noise to be constant, but usually the system noise vary over time, which causes interference to the parameter filtering. In this paper, we introduce adaptive kalman filter based on maximum likelihood estimation to estimate and remove the system noise in real time. This method can capure a better true movement and get a better image stabilization effect.Experiment results show that in complex jitter scenes our algorithm is better than FPS, Lee's method and Deshaker. For the other videos, the result of our algorithm is close to other three methods. So this algorithm could give a very good solution for video stabilization challenges.