Noisy MPEG Motion Vector Reduction for Motion Analysis

In MPEG standards, motion compensation is used for inter-frame compression. Motion compensation generates motion vectors in order to predict the current frame regions from previously decoded frames. These motion vectors represent motion information between regions in different frames and are useful in motion analysis. However, the motion vectors of homogeneous, low-textured, and line regions tend to be unstable and noisy; therefore, it becomes difficult to conduct motion analysis. In this paper, we propose a noisy motion vector reduction method for motion analysis using MPEG motion vectors by introducing a global motion estimation method and a zero-comparison method. The proposed method outputs local motion vectors that are free of noise and high-quality global motion vectors that can be used for object detection, tracking, etc. Further, this method works even for videos captured by a moving camera. We demonstrate the effectiveness of the proposed method through several experiments using actual videos.