Single Camera Path Detection for Outdoor Navigation

Ondrej Jariabka, Marek Šuppa, Ondrej Rudolf

Supervisor(s): Mgr. Pavel Petrovič, PhD.

Comenius University

Abstract: The ability to find a path leading to an objective is one of the fundamental characteristics of autonomous mobile robots. In the recent years approaches which make use of a combination of Computer Vision and Machine Learning have gained traction in the robotic community. In this work we examine the problem of Path Detection in the context of the Robotour competition by analyzing proposed solutions that use the aforementioned combination. Furthermore, we propose new models based on recent advances in using Convolutional Neural Networks for related Computer Vision tasks as well as structural changes to the processing pipeline, which help to improve its performance. Finally, we evaluate all of the described methods on a dataset that is released as part of this work.
Keywords: Computer Vision
Full text:
Year: 2017