Optical flow computation via multiscale regularization

Abstract

Abstract : The apparent motion of brightness patterns in an image is referred to as the optical flow. In computational vision, optical flow is an important input into higher level vision algorithms performing tasks such as segmentation, tracking, object detection, robot guidance and recovery of shape information. In addition, methods for computing optical flow are an essential part of motion compensated coding schemes. In this paper, we present a new approach to the problem of computing optical flow. Standard formulations of this problem require the computationally intensive solution of an elliptic partial differential equation which arises from the often used regularization term. We utilize the interpretation of the smoothness constraint as a fractal prior to motivate regularization based on a recently introduced class of multiscale stochastic models. These models are associated with efficient multiscale smoothing algorithms, and experiments on several image sequences demonstrate the substantial computational savings that can be achieved through their use.