Identification of human movement through video processing

Abstract

Studies of face recognition and movements of the human body have opened frontiers for different knowledge areas. The objective of this paper is to show an alternative methodology to identify faces and make a characterization of the human body movement. This methodology has three phases. The first phase is a generation of a video database in a NoSQL model using MongDB. The second phase permits to identify human faces applying an algorithm that combines Hough Transform and the Viola-Jones algorithm. And the third phase makes an identification of human movement using optical flow histogram-oriented gradients. The main results show that the face identification permits to select a relevance range for an approximation of the interest or actor area. This interest area is a base to apply the optical flow, avoiding false positives, like shadows or reflections produced by the environment. Finally, we conclude that the obtainment of a clear frontal view of the face supplies an easy way for human action tracking.