Undergrad Research Project - Single Camera SLAM

Spring 2017

Vincent Liu
Kris Kitani
Project description

The objective for this research project is to develop a robust, lightweight, and real-time single camera Simultaneous Localization and Mapping (SLAM) system. I will look into the area of dense subpixel alignment of binary descriptors employing hamming distance. This usage of these image alignment features will hopefully improve a single camera SLAM system to less than 1 meter accuracy within well-defined environments. I will also be looking at how the usage of these low-complexity dense binary descriptors, known as bit-planes, perform in various conditions and aspects of a novel three parallel thread SLAM system. This will make use of various concepts such as a spanning tree and a covisibility graph. The anticipated results are to eventually get an extremely accurate and real-time SLAM system for use on an iPhone. This would then enable the phone to "see" the world around it, being able to localize in a map, track movement, navigate, along with many other possibilities.

Return to project list