Research On Techniques For Video Motion Information Analysis

In recent years, motion and video computing has become a hot topic in the domain of computer vision. In addition to those traditional visual motion analysis based on video se-quence, such as computing optical flow (2D motion) and structure from motion (3D motion and shape), the motion information presented in a video sequence is also being used to solve several other problems, such as video synthesis, video segmentation, video compression, video registration, and video surveillance and monitoring.A video sequence possesses not only spatial properties like images, such as color, edge, texture, but also temporal properties, namely, the motion information. Thus, compared with the traditional image analysis techniques, motion information analysis based on video se-quence plays more important role in solving the above mentioned problems. However, in the practical applications, the moving background caused by camera motion, jittering camera, foreground objects'irregular motion etc make the video motion information analysis a tough problem. To solve the problem, a thorough research on the video motion information analysis has been carried through in this dissertation. The main contributions are as follows:1. Fast and accurate global motion estimationGlobal motion estimation (GME) is to estimate the law of camer motion, which causes the background moving. By GME, the pixel correspondences between adjacent frames can be obtained, and thus the problem with complex moving background can be simplified to the one with static background. Therefore GME is the fundament for the motion information analysis in video sequences with dynamic background.An adaptive outlier filtering algorithm is proposed and used to improve Konrad's GME method in MPEG-4 validation model. By estimating global motion parameters and adaptively filtering the outliers alternately, the noise is eliminated effectively. Compared with Konrad's method, our method improves the speed of GME nearly 1 times, while the accuracy of the GME is guaranteed.2. High-quality stabilization of shaky videosVideo stabilization is to remove the visually annoying vibrations in videos caused by un-wanted camera motion. As vibrations in videos will not only cause the degradation of the visual quality, but also make the further analysis more difficult, so the stabilization of shaky videos is necessary.A video stabilization algorithm based on multi-trajectory mapping is proposed. By map-ping the estimated camera motion to multiple trajectories and smoothing the trajectories, the unwanted camera motion is eliminated. Then, according to the parameters of the observed camera motion and the smoothed camera motion, video frames are reconstructed to obtain a stabilized video with high visual quality. Compared with current stabilization algorithms, our...