Hardware and software considerations for monocular SLAM in a riverine environment

A monocular SLAM algorithm for use on rivers is proposed and compared to existing methods using a newly created SLAM dataset. The proposed algorithm uses a single camera and inertial measurements to estimate the location of a canoe and a map of a river simultaneously using an extended Kalman filter. The algorithm exploits the reflections of map landmarks in the river in order to obtain a depth estimate from a single view. Landmark reflections are found by using the state covariance matrix of the extended Kalman filter to define a search region where reflections are likely to be found. A process noise model is proposed to more accurately reflect the noise characteristics of the inertial measurement unit. The dataset used for the experiments was collected from a canoe on the Sangamon River covering 2.7 kilometers in 44 minutes and divided into eight subsets. Data collected includes stereo images, inertial measurements, and GPS position data for ground truth. The proposed algorithm is evaluated by measuring the translation and attitude error with respect to ground truth and comparisons are made to the stereo method, ORB-SLAM2.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2018

Author dc:creator

• Miller, Martin Hudson

Contributors dc:contributor

• Hutchinson, Seth A.
• Chung, Soon-Jo

• Simultaneous localization and mapping (SLAM)
• Monocular
• Stereo
• River
• Inertial measurement unit (IMU)
• Camera
• Robotics
• Inertial

Statement dc:rights

• Copyright 2017 Martin Miller

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/99412

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/99412