Monocular vision based navigation using image moments of polygonal features

This thesis presents a novel monocular-vision-based localization and mapping algorithm using moments of polygon features. The landmarks we use are polygonal regions instead of a dense set of feature points, which can significantly reduce the computational complexity of data association and produce a map that is geometrically and structurally more meaningful. Each region can be characterized using its depth and orientation with respect to the camera and an polygon detection and tracking algorithm is developed. The monocular vision Simultaneous Localization and Mapping (SLAM) problem is formulated as a filter problem to incorporate the image moments of the close regions or polygons tracked. The observability of the SLAM estimator is further improved by both the additional measurements with respect to the initial view location and the use of image moments. We analyze the performance of our SLAM algorithm with numerical simulations and experimental results. We also compared our results with ORB-SLAM to show the effectiveness of our algorithm in outdoor environments.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Mechanical Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2017

Author dc:creator

• Ma, Lingyu

Contributors dc:contributor

• Hutchinson, Seth
• Chung, Soon-Jo

• Simultaneous localization and mapping (SLAM)
• Vision-based navigation

Statement dc:rights

• Copyright 2016 Lingyu Ma

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/95582

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/95582