A Software Architecture for Distributed Visual Tracking in a Global Vision Localization System

The paper considers detecting and tracking multiple moving objects in real time by a multiagent active vision system. The main objective of the envisioned system is to maintain an explicit comprehensive representation of the scene by combining individual views obtained from several autonomous observer nodes. In order to allow for a near real time performance, each observer agent has been assigned a separate computer system, while an additional coordination agent is introduced for relieving the observers from correspondence and coordination tasks. The proposed architecture is specially suited for monitoring ground objects whose dimensions are relatively small when compared to the dimensions of the scene. This assumption makes it possible to speculate a ground object 3D position from the single view, which consequently allows a robust correspondence approach. The architecture has been implemented in an experimental global vision system, whose final objective is to provide localization information for a group of simple robots without vision sensors. The system was preliminary tested in the laboratory environment, and the obtained experimental results are presented.