Quasi-dense optical flow estimation based on dynamical system models

Optical flow is a vector which represents the motion of objects in images. In this paper, new methods for dense and quasi-dense optical flow estimation are proposed. The dense estimation method can be applied to optical flow estimation at each pixel in an arbitrary area. The proposed dense method considers an optical flow equation at each pixel as a dynamical system. Robust estimation can be achieved with dense intensity information, especially against illumination changes. In order to improve estimation performance at zero optical flow, a quasi-dense estimation is also proposed.