A novel approach for finding the movement of an object in video sequences by an artificial neural network for 2.5D object modeling

In this paper we propose a system that identifies and tracks the movement of an object appearing fully or partially hidden by occlusion in a video sequence for the ultimate purpose of modeling the moving object in 2.5D space using the SfM (structure from motion) concepts. This paper presents a novel algorithm to detect moving objects in video sequences by first performing image segmentation on the frame sequences based on the criteria of motion, and then applying a motion vector estimation algorithm to find geometrically identical points in two consecutive video frames. An ANN (artificial neural network) based model was adopted to segment the moving object(s) out of the stationary background. The next step involves applying motion vector search on the motion segmented images to obtain a correspondence between a pixel of the object in the reference frame and a pixel in the subsequent frame such that the pixels corresponds to the same part and geometrical location of the object. Results from various video sequences of motion based segmentation using ANN and the subsequent motion vector estimation have been presented in this paper. Eventually, a wire-frame diagram is constructed to represent a moving object in 2D