Simultaneous Localization and Mapping (SLAM) in a single camera system
This paper discusses a way of solving the Simultaneous Localization and Mapping (SLAM) problem using only a single camera system. While this is challenging because of limited amounts of information available from a single camera, it provides benefits in terms of cost, weight, and speed of processing. The paper starts by breaking the problem down into three topics: target/video tracking, camera calibration/mapping and probabilistic filtering. For each of these topics, a suitable method is decided upon and an algorithm implemented in Python. The algorithms are tested on a combination of dataset videos as well as simulated scenarios and the quality of the algorithms discussed. Issues and challenges for each section are explored and discussed, and the algorithms iterated on to create a system that is effective under difficult scenarios. The result is a high-quality video tracking algorithm and SLAM system that works well using simulated data. Reasonable extensions to the work are suggested to develop upon the findings.
