A control theoretical approach to active vision

Abstract

Integration of active vision processes is a primary problem for construction of systems that have a robust performance. Active vision systems may be broken up into fixation, pursuit, and attention processes. These processes have very different spatio-temporal scope. If the scope is used as a guiding principle for organizing a system, a hierarchical architecture can easily be formulated. The integration strategy can in addition be based on traditional control methods, where different error/position estimates are integrated using simple summation between the control loops. In this paper we have demonstrated how image based processes, 2D and 3D tracking processes may be integrated into a coherent system. A system that is based on these principles has been designed and implemented. The system can operate at a speed of 7 Hz is fairly general settings. Experiments with the implemented system has shown that the performance of such a system is superior to the traditional 2D tracking systems.