Abstract
Visible light communications (VLC) using LED lights as transmitters is an emerging technology that can provide services such as illumination, localization, and mapping, in addition to communications. Since the visible light spectrum is unregulated, and the cost of VLC systems is decreasing, the widespread deployment of VLC systems is imminent.
In this dissertation, we begin by introducing a powerful VLC-based indoor positioning method that uses probabilistic tracking filters and a pre-collected database of optical power distribution, also known as a fingerprint map. In fingerprint-based localization methods, the location-related data collection process needs to be done a priori. This makes fingerprint-based localization algorithms difficult to use. We propose three ways of automating or progressively building the fingerprint map. In the first method, we utilize methods from geological science that are used to build a map of a surveillance area, namely, ordinary Kriging and radial basis functions. In the second method, we propose to equip the lamps with cameras to capture the light intensity distribution. The fingerprint map can be extracted from these images. In the final method, we rely on a large number of users that are connected to Wi-Fi to collect the light intensity measurements for us. We test the effect of these different data collection methods on agent tracking using Kalman and particle filters and find that the localization accuracy remains on the order of the fingerprint map resolution despite collected data inaccuracies. We also investigate the effect of unexpected failures in the VLC infrastructure on agent tracking. The root mean square error (RMSE) of VLC-fingerprint tracking is around 6 centimeters for a fingerprint grid step size of 10 centimeters.
To achieve accurate localization, prior belief in the landmark locations must be absolute. In reality, the landmark locations may not be given a priori. We use a distance geometry-based method to localize the LED landmarks. The method does not require any prior information about either the LED or the agent location. The resulting agent localization accuracy decreases proportionally to the landmark localization error. The landmark localization RMSE is better than 20 centimeters for our proposed method, and the agent tracking RMSE is around 10 centimeters.
After localizing itself, an agent needs to learn its environment. Indoor mapping is a problem that has garnered much attention, and the widely accepted solution is to use LIDAR. Although LIDAR provides excellent maps with high fidelity, they are bulky, expensive, and signal processing intensive. In this dissertation, we propose a sparse indoor mapping method that requires lightweight signal processing and inexpensive sensor design. We utilize the channel state information (CSI) of the optical wireless communication system and obtain the time-of-flight of the reflected light rays from the CSI. The technique is able to generate a coarse outline of the indoor space with respect to the agent position. With the use of an additional outlier rejection step, the mapping RMSE reduces to less than 5 centimeters at high signal-to-noise ratio values.
