Visible light communications (VLC) based on illuminative LED lamps are attracting increasing attentions due to its numerous advantages. Compared with conventional lamps, LED illumination sources have long operational life, low power consumption, and are mechanically robust. Therefore LEDs are expected to be the next generation of lamps that will be widely installed to replace conventional lamps. Besides, LEDs can be modulated at very high speeds, which allows the possibility of simultaneously providing communication while illuminating. The light modulation frequency is sufficiently high that it is undetectable by the human eye, yet detectable by arrays of photodiodes. Considering these facts, a system can be designed to receive (using a single photodiode or a camera) and analyze LED signals. Furthermore, such a system should be able to facilitate position estimation tasks either for people or vehicles.
Two kinds of sensors - single photo-detector and array photo detector (camera or linear array) C can be used to detect the LEDs. The single photo-detector approach is used to detect the existence and the specific identity of an LED. The array photo-detector approach measures the angle-of-arrival of the LED signal. A single photo detector provides the simplest hardware approach and could communicate with LED's at a very high data rate, but only offers the most basic level of positioning. Cameras provides much more informative position measurements; however, there are challenges to achieving high rate LED-to-camera data communications due to the current hardware architectures of cameras. Alternatively, linear PD arrays allow high sample rates, high accuracy and low cost. This dissertation will investigates the issues (observability, extrinsic parameter calibration, and vehicle state initialization) related to implementation of a positioning and communications system built on a linear optical sensor array.
The VLC based navigation methods require recovering the LED ID from a sequence of photo-detector scans. This ID will help the data association in the navigation system or the data communication in the VLC system. Recovering LED data (ID) requires identifying each LED's projection position and on-off status in each photo-detector scan. Identifying the LED projection is challenging because: 1) Clutter and noise corrupt the measurements; 2) The LED status will be "off" in some frames; 3) The predicted projection location sequence depends on the estimated vehicle state trajectory, which is uncertain. This dissertation presents two new methods determining the most probable data and LED position sequences simultaneously, using Bayesian techniques by maximizing posterior probabilities. The first method is based on Viterbi algorithm and developed under the assumption that the frame rate of the array sensor is high relative to the rover motion bandwidth. The second one is based on multiple hypothesis tracking (MHT) techniques with no assumption assumed. Both the two methods are analyzed theoretically and demonstrated by experimental results.