A PC-Cluster based Multi-Camera System for Real-Time Multiple Objects Tracking

Abstract

This paper presents a PC-cluster based multi-camera system for real-time multiple objects tracking. The PC-cluster, a group of linked computers, is developed by using PCI-to-PCI data mover card to improve real-time processing performance and provide system scalability. Five processing modules have been designed and implemented to control system operations. Target objects are in spherical shape. Four IEEE-1394a cameras with hardwired synchronization are used for system testing. After identifying correspondence objects in different views, 3-D coordinates of each target have been reconstructed. Experimental results show that the system can identify and track multiple objects moving in 3-D space correctly. The maximum frame rate for tracking is 12 frames per second. In the near future, the system can be applied for specific application when additional processing modules are added to the system