Contour Detection of Multiple Moving Objects in Unconstrained Scenes using Optical Strain

Abstract

Moving Object Detection (MOD) is still an active area of research due to the amount of scenarios it can tackle and the different characteristics that may appear in them. Therefore, getting a unique method that performs well in all the situations becomes a challenging task. In this paper we address the MOD problem from a physical point of view: given the optical flow between two images, we propose to find its motion-boundaries by means of the optical strain, which gives information about the deformation of any vector field. As optical strain detects all the motions from a sequence, we propose to work on temporal windows and apply thresholding on them in order to separate noise from real motion. The proposed approach shows competitive results when compared to other methods on known datasets.