Loading...
Date of Award
6-2-2018
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Systems Engineering
First Advisor
Xian Liu
Abstract
For a visually impaired individual, it is a challenging task for him/her to plan and follow a path towards the destination. This task is referred to as wayfinding. In this paper, the functions of wayfinding are defined as localization (aka pose estimation) in an indoor environment and finding a way to get to the destination by using the location information. To address this issue, vision-based navigation systems that use a camera for pose estimation have been intensively studied. Several vision-based robotic navigation aids (RNAs) for the visually impaired have been developed. However, these RNAs are not reliable enough to assist the visually impaired for wayfinding because of featureless scenes, occlusions, abrupt motion and illumination changes. The objective of this dissertation is to develop a robust 6-DOF pose estimation method for a particular RNA—Co-Robotic Cane(CRC)—for wayfinding for the visually impaired. The main contributions of this research are highlighted as follows: First, a new pose estimation method that uses the geometric information of the operating environment (extracted from the range data of a 3D time-of-flight (ToF) camera) to reduce accumulative pose error is proposed. Based on the method, an indoor wayfinding system is developed and validated by experiments. In addition, the developed RNA has been tested by human subject experiments. Second, a new factor graph based multimodal data fusion algorithm is proposed to integrate visual information and range data from a 3D camera and the inertial data from an inertial measurement unit (IMU) for robust pose estimation in indoor environment. Pose estimation by coupling the measurements from a camera and an IMU is termed as visual-inertial odometry (VIO) or visual-inertial SLAM (VI-SLAM). To improve the existing VIO approaches’ accuracy, a new method, called plane aided visual inertial odometry (PAVIO), is proposed. The method uses plane features of the operating environment to identify accurate VO outputs to improve VIO pose estimation accuracy. Third, the suitability and performances of three state-of-the-art tightly-coupled VI-SLAM methods are investigated and compared in the context of CRC navigation. Based on the results, the most suitable method is selected and extended for wayfinding, enlightening the future improvement for RNA development
Recommended Citation
Zhang, He, "Robust Pose Estimation of a Robotic Navigation Aid for the Visually Impaired by Multimodal Data Fusion" (2018). Theses and Dissertations. 820.
https://research.ualr.edu/etd/820
