Project dates and setting
The project ran from August 2012 to May 2014, the period of Dr. Costilla Reyes’s master’s studies in electrical engineering at the University of North Texas (UNT). As a graduate research assistant in the College of Engineering he assisted with research funded by the National Science Foundation, NASA, and other federal agencies, and his master’s research spanned wireless sensor networks, machine learning, computer vision, and perception for autonomous systems and robotics. In spring 2013 he gave a project presentation showing a map and localization result for the hallway of the electrical engineering building.
What the system did
Simultaneous localization and mapping (SLAM) is the problem of building a map of an unknown space while estimating the robot’s own position within it, so that neither can be assumed in advance. The RGB-D (color plus depth) approach used here relied on a Kinect camera, which supplies a depth value for every pixel alongside ordinary color video. Matching visual features across frames and aligning the associated depth points lets the robot track its own motion and accumulate a three-dimensional model of the indoor environment. Dr. Costilla Reyes implemented the method and optimized it to run efficiently on live video from the robot.
The hallway result from the 2013 presentation shows what such a method produces: a colored three-dimensional point cloud of the corridor walls, with the estimated camera path traced through it. The result illustrates the method from end to end, from camera frames to a map and an estimate of the robot’s position within it.
A related approach to indoor positioning
His master’s thesis addressed the same underlying question in a different way: how to estimate position indoors, where satellite navigation is unavailable. The paper from the thesis, on dynamic Wi-Fi fingerprinting for indoor positioning, was presented at the 2014 International Conference on Indoor Positioning and Indoor Navigation. The SLAM project used camera depth data, while the fingerprinting work used wireless signal measurements.