Toward Persistent Tracking and Identification in Camera Sensor Networks

A Dissertation submitted in partial satisfaction
of the requirements for the degree of
Doctor of Philosophy
Electrical Engineering
Michael James Quinn


In recent years, research in the area of camera sensor networks has accelerated dramatically with the increased availability of cheap sensing, processing, and communications hardware. Design, implementation, and most importantly the operation of camera networks provide numerous challenges for vision researchers. The first challenge encountered is usually the implementation of a test system in which research can be performed. We provide a detailed overview of our work on the VISNET system. The VISNET system is a ten-node vision testbed located in UCSB's Harold Frank Hall. The system is composed of standard off the shelf hardware, utilizing PCs and IEEE 1394 cameras. The software is developed using freely available resources, including OpenCV and ffmpeg. We present two applications in the VISNET system: distributed network calibration and multicamera tracking. We then approach the problem of sensor selection in a camera network. We first present a scoring system for selecting camera nodes for localization and tracking. We then extend this system to minimize the number of node activations and handoffs during the tracking process. Second, we present a view scoring system for multiview appearance model learning in camera networks. The system collects the best views from several poses of a tracked person and uses them to assemble a model which captures appearance variation as a function of view angle. In summary, we present work which advances the current state of camera networks research by providing guidance on both test system construction and system operation.

[PDF] [BibTex]
Michael James Quinn,
Ph.D. Thesis, University of California, Santa Barbara, Dec. 2008.
Node ID: 520 , DB ID: 327 , Lab: VRL , Target: Thesis