E. G. Ortiz, J. Schiff, K. Goldberg."Sensitivity Analysis for Intruder Tracking Using Particle Filtering and a Network of Binary Sensors". In the Proceedings of SUPERB-IT. Department of EECS, August, 2006.
I worked on this project in the Berkeley Laboratory for Automation and Science Engineering during a summer, undergraduate research experience. I worked under the mentorship of Jeremy Schiff and Dr. Ken Goldberg.
Our lab has designed a security system to automatically track and capture photos of an intruder using a high-resolution robotic pan-tilt-zoom camera using a wireless network of inexpensive binary sensors. The sensors suffer from a refractory period where they may be unresponsive. An estimation method based on Particle Filtering, a numerical sequential Monte Carlo technique, takes the data from these noisy sensors and produces a predicted path. The sensors are modeled with conditional probability density functions and incorporate a probabilistic model of intruder velocity. This paper assesses the sensitivity of the system to a number of design parameters. These include the resolution of the world-space representation, the number of samples used to represent the distribution of the object’s location, different sensor classes, and sensor placement. This work provides an understanding of which choices have greatest impact to the system’s performance. Due to the real-time constraints, there is only a finite amount of computation available per iteration. We also investigate the tradeoff between the complexity of sensor/intruder modeling and the frequency of processing the data.