Quasi-motion extraction utilizing edge field analysis

Recognizing the motion in the images is an important part of computer vision. In order to recognize the motion in the images, it is significant to distinguish three motion primitives as translation, rotation and divergence that can be extracted based on the optical flow. In this paper, we introduce a method for extracting quasi-motion utilizing edge flow in one image, namely edge field analysis. In general, the optical flow is estimated utilizing two or more successive images. On the other hand, this method directly estimates translation, rotation and divergence not utilizing two or more images but utilizing only one image without the information of the optical flow of the motion. We firstly formulate the problem of extracting the motion primitives as a vector field analysis of the velocity vector field. We then propose an algorithm for estimating three motion primitives from one image regarding edge as pseudo-velocity vector. The experimental results in terms of estimating the eye of the storm, Whirlpool Galaxy and the center of the explosion are also demonstrated to evaluate the proposed method.