Abstract

Visual odometry is a challenging approach to simultaneous localization and mapping algorithms. Based on one or two cameras, motion is estimated from features and pixel differences from one set of frames to the next. A different but related topic to visual odometry is optical flow, which aims to calculate the exact distance and direction every pixel moves in consecutive frames of a video sequence. Because of the frame rate of the cameras, there are generally small, incremental changes between subsequent frames, in which optical flow can be assumed to be proportional to the physical distance moved by an egocentric reference, such as a camera on a vehicle. Combining these two issues, a visual odometry system using optical flow and deep learning is proposed. Optical flow images are used as input to a convolutional neural network, which calculates a rotation and displacement based on the image. The displacements and rotations are applied incrementally in sequence to construct a map of where the camera has traveled. The system is trained and tested on the KITTI visual odometry dataset, and accuracy is measured by the difference in distances between ground truth and predicted driving trajectories. Different convolutional neural network architecture configurations are tested for accuracy, and then results are compared to other state-of-the-art monocular odometry systems using the same dataset. The average translation error from this system is 10.77%, and the average rotation error is 0.0623 degrees per meter. This system also exhibits at least a 23.796x speedup over the next fastest odometry estimation system.

Library of Congress Subject Headings

Odometers; Computer vision; Motion detectors; Neural networks (Computer science)

Publication Date

11-2016

Document Type

Thesis

Student Type

Graduate

Degree Name

Computer Engineering (MS)

Department, Program, or Center

Computer Engineering (KGCOE)

Advisor

Andreas Savakis

Advisor/Committee Member

Raymond Ptucha

Advisor/Committee Member

Roy Melton

Comments

Physical copy available from RIT's Wallace Library at TL272.5 .M85 2016

Campus

RIT – Main Campus

Plan Codes

CMPE-MS

Share

COinS