Dynamic textures segmentation and tracking using optical flow and active contours

Abstract

A large number of scenes composing our visual world are perceived as dynamic textures, displaying motion patterns with a certain spatial and temporal regularity such as swaying trees, smoke, fire, human movements, flowing water and others. In real scenes, encountering dynamic texture superimposition is quite frequent, in which, we are challenged to separate each region aside in order to improve their analysis. This research paper presents a novel approach for segmenting then tracking dynamic textures in video sequences, using optical flow and static manual active contours, which we adapt to be dynamic and fully automatic. Experiments were conducted on DynTex and YUP++ datasets, where the achieved results demonstrated a success of the proposed approach to segment and track dynamic textures effectively.