A Dense Optical Flow Estimation Based on Dynamical System Approach under Intensity Variation Violating the Conservation Law

Optical flow is a vector that represents the motion of objects in images. To estimate the optical flow, most of the optical flow estimation methods assume the intensity conservation law. The estimation methods based on a dynamical system approach [3], [4] also use the same assumption. Therefore, these methods might not be able to estimate correctly if the intensity conservation law is violated by various factors. A estimation model at a pixel considering the illumination change was proposed in [7] to improve the estimation accuracy in such situations. This paper proposes a new dense estimation model with the intensity correction quantity to overcome the estimation in the illumination change and the occlusion. Furthermore, this paper proposes to estimate the optical flow with the Kalman filter or the Kalman filter with the sparse constraint to correction quantity. Using the sparse constraint, the correction quantity becomes non-zero only near the flow boundaries. Accordingly, the improvement of estimation accuracy can expect. The effectiveness of the proposed models is demonstrated through numerical examples.