Efficient human pose tracking using multiplex sensing
We design and calibrate an efficient human detection system, capable of detecting and tracking a single person while minimizing the number of required sensors. Our infrared detectors have a 2 meter range and a 250ms rise/fall time. We sample their signal using an MSP430F149 microprocessor, correlate the results to reference patterns, and collect the binary decisions of each sensor on a workstation wirelessly. We modulate the optical field of the detectors by introducing selectively opaque reference structures into their optical path. Segmenting the physical space into signature cells enables direct measurement of source configuration. Brute force determination of the mapping is prohibitively expensive; we propose a method to estimate this mapping and predict signatures for every source position by observing signatures along prescribed tracks through the physical space.
