Date of Award

6-10-2015

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Systems Engineering

First Advisor

Cang Ye

Abstract

The objective of this dissertation is to develop a 6-DOF pose estimation method for a portable navigation device, called Smart Cane (SC), for the visually impaired. The SC is a computer-vision-enhanced white cane that uses a single 3D time-of-flight camera for both device localization and object/obstacle detection. The 6-DOF device pose is computed by two steps. First, a Pose Change Estimation (PCE) method, called Visual Range Odometry (VRO), determines the Pose Change (PC) between two camera views by tracking visual features over the two image frames. This PC information is then integrated over time to obtain the device pose in the world coordinate. To retain necessary accuracy and repeatability of PCE in various real-world environments, a filtering method is applied to the camera's data to reduce the noise-induced error and a Fast Iterative Closest Point (FICP) method is proposed to avoid the VRO performance degradation in visual feature-sparse environments. The FICP method refines the VRO result by aligning the 3D data points that are common to the two camera views. To reduce the SC's pose error growing over time during the pose integration process, a pose graph is constructed using the PCs computed by the FICP-enhanced VRO method and a Pose Graph Optimization (PGO) method is applied to minimize the pose error for each node. Finally, a computationally efficient Loop Closure Detection (LCD) method is introduced to reduce the SC pose error when the 3D camera revisits a same place due to the swing motion of the SC along the path. The method is also used to detect the long-term loop closure for a looped trajectory and further reduce the SC's pose error. The proposed VRO, FICP-enhanced VRO, PGO and LCD methods are validated with datasets collected from real environments with the SC. The results show that the proposed methods improve the pose estimation accuracy and reduce the computational cost. Eventually, a wayfinding system prototype, including hardware and software, is built and used to demonstrate the use of the pose estimates for wayfinding. The experimental results show that the entire system functions correctly and provides the location information and wayfinding commands to its user in an indoor environment.

Download

Share

COinS