Camera-based localization for indoor multi-robot system experiments

Abstract

Robotics has become one of the most popular and promising fields in computer science nowadays, but the development takes a lot of effort. One of the most important aspects of robotics is localization, which means the act of finding the position and direction of a robot or other object. Localization is especially important when developing autonomous mobile robots, as even the tiniest errors in tracking can add up to huge errors. To properly test tracking methods during development a way to extract the ground truth is required, and there are multiple ways to achieve this. Probably the most well-known example of this is the Global Positioning System, or GPS. However, GPS is ineffective indoors due to weakened signal, and the precision is not suitable for accurate room-scale positioning.

This paper focuses on comparing different approaches for obtaining the ground truth positional data for mobile ground robots using a ceiling-mounted video camera. The desired features for the methods include (but are not limited to) real-time tracking, ability to locate multiple robots, and overall robustness against changes in environment like obstacles and lighting changes.