Date of Award
Master of Engineering Science
Mechanical and Materials Engineering
Dr. Michael D. Naish
Dr. George K. Knopf
This thesis introduces a centralized active vision system for the remote identification of multiple targets in applications where the targets may outnumber the active system resources. Design and implementation details of a modular active vision system are presented, from which a prototype has been constructed. The system employs two different, yet complimentary, camera technologies. Omnidirectional cameras are used to detect and track targets at a low resolution, while perspective cameras mounted to pan-tilt stages are used to acquire high resolution images suitable for identification. Five greedy-based scheduling policies have been developed and implemented to manage the active system resources in an attempt to achieve optimal target-to-camera assignments. System performance has been evaluated using both simulated and real-world experiments under different target and system configurations for all five scheduling policies. Parameters affecting performance that were considered include: target entry conditions, congestion levels, target to camera speeds, target trajectories, and number of active cameras. An overall trend in the relative performance of the scheduling algorithms was observed. The Least System Reconfiguration and Future Least System Reconfiguration scheduling policies performed the best for the majority of conditions investigated, while the Load Sharing and First Come First Serve policies performed the poorest. The performance of the Earliest Deadline First policy was seen to be highly dependent on target predictability.
Ward, Christopher D. W., "Development of an Active Vision System for the Remote Identification of Multiple Targets" (2008). Digitized Theses. 4136.