Agents and Computer Vision for Processing Stereoscopic Images

Abstract

This paper presents a Multi-Agent System (MAS) that implements techniques of Computer Vision for processing stereoscopic images by using stereo cameras. The MAS focuses on detecting people and their behavior through a two-phase method. In the first phase, the MAS creates a model of the environment by using a disparity map. It can be constructed in real time, even if there are moving objects in the area (such as people passing by). In the second phase, the MAS is able to detect people and their behavior by combining a series of techniques such as Sum of Absolute Differences (SAD) or Gradient Orientation Histograms (HOG). The preliminary results and conclusions after several experiments performed on real scenarios are described in this paper.