View source for DTU running path dataset


== Vision Assisted Path Navigation for Autonomous Robot ==

Frederik Emil Thorsson Schöller (s133910)
Martin Krarup Plenge-Feidenhans’l (s125097)
Master of Science in Engineering
2019

Full thesis: https://findit.dtu.dk/en/catalog/2444130462

=== Abstract ===

The main focus of this project is to propose a solution enabling a robot to robustly
navigate a chosen forest path utilizing computer vision. The robot should be able to
navigate the entire path, excluding public roads. While autonomously navigating the
path, the solution should enable the robot to avoid obstacles such as pedestrians while
also providing means to stop the robot when approaching public roads.
A deep learning approach is used, providing the solution with a semantic segmenta-
tion of the acquired images, splitting these into path and background. The segmented
images are processed using classical image analysis in order to provide the robot with
a driving direction, making it follow the path. A GPS based point of interest system
has also been added, enabling the robot to stop when approaching traffic, while also
providing a way of e.g changing parameters when driving a particularly difficult path
segment.
Through testing, the developed system has been successfully evaluated and the sug-
gested solution was found to fulfill the project goals.

[[File:path_segmentation_result.png]]

=== Images and annotations ===

The images and annotations (8679 images) - in total 2.6 GB

http://aut.elektro.dtu.dk/staff/jca/running_path_data_set.tar.gz 

=== Source code ===

Python scripts used for the project

http://aut.elektro.dtu.dk/staff/jca/path_navigation_source.zip