An indoor positioning system facilitated by computer vision

Abstract

The purpose of this paper is to present our work on a novel method which allows for previously unattainable accuracy in indoor positioning. Global positioning has changed the way in which we interact with our specific locations on a real time basis, as can be seen most prominently in mapping applications. However, global positioning is severely limited indoors where location is equally important, and further, requires greater accuracy. While there have been attempts to implement indoor positioning, these methods are severely lacking, prompting us to take a completely new approach. We use low cost webcams and a series of algorithms to detect people in a video frame, and then identify and position them. Accuracy for identification is upwards of 95% and positioning accuracy is within a half-meter for the majority of the frame of view, all while running in real time on mobile CPUs. Such a system can be implemented on large scales to allow for exciting new applications; indoor directions in malls and public transportation hubs, new forms of human-robot interactions and consumer habit analysis in stores are all now possible.