Wireless Sensor Networks for Spectral Camera Calibration

Sunayana Saha

M.S. dissertation, Department of Computer Science, University of Illinois at Urbana-Champaign, 2003

Peter Bajcsy, Advisor

This work presents a unique spectral camera calibration technique using wireless "smart" micro electro-mechanical (MEMS) sensors. The word "smart" denotes capabilities of MEMS other than sensing, for instance, computing and communication. We foresee the use of widely distributed and deeply embedded "smart" MEMS sensors as potential calibration gauges for spectral cameras, such as, thermal infrared (IR) or visible spectrum cameras. Thus, the primary motivation of our presented work is to investigate the problem of spectral camera calibration in an indoor environment using the "smart" MEMS sensors. The main application of our work is in hazard-aware environments, or in any smart and intelligent spaces. There are also many other applications of advanced computer vision and robotics that try to automate manufacturing processes and can benefit from the use of "smart" MEMS and the results of our work.

We have investigated an optimal design of a spectral calibration system with thermal IR and hyperspectral cameras and the "smart" MEMS sensors known as the MICA sensors.

The MICA sensor hardware is manufactured by Crossbow Inc. and programmed using TinyOS, an open source operating system, developed by the University of California at Berkeley. Our experimental results demonstrate a calibration of the thermal IR camera manufactured by Indigo System Corporation with MICA temperature sensors and our preliminary calibration of MICA luminance sensors using a hyperspectral camera manufactured by Opto-Knowledge Systems Inc.

In this thesis, we provide an overview of the features and challenges of the MEMS sensors in general, as recognized by current research efforts. We then propose a robust calibration system design by maintaining two main objectives ñ minimizing the wireless loss of data transmitted by the sensors and maximizing the information content of the collected data. The first objective is fulfilled by experimentally determining an optimal sensor network that gives the least data losses. Maximization of information content is achieved by synchronizing, registering and calibrating acquired 2D spectral camera images with precise point measurements obtained wirelessly from the MICA sensors. We believe that the reported outcomes of our experimental results will help other researchers designing optimal spectral camera calibration systems in future.