Defense Date


Document Type


Degree Name

Doctor of Philosophy


Computer Science

First Advisor

Cang Ye


This dissertation aims to address the limitations of existing visual-inertial (VI) SLAM methods - lack of needed robustness and accuracy - for assistive navigation in a large indoor space. Several improvements are made to existing SLAM technology, and the improved methods are used to enable two robotic assistive devices, a robot cane, and a robotic object manipulation aid, for the visually impaired for assistive wayfinding and object detection/grasping. First, depth measurements are incorporated into the optimization process for device pose estimation to improve the success rate of VI SLAM's initialization and reduce scale drift. The improved method, called depth-enhanced visual-inertial odometry (DVIO), initializes itself immediately as the environment's metric scale can be derived from the depth data. Second, a hybrid PnP (perspective n-point) method is introduced for a more accurate estimation of the pose change between two camera frames by using the 3D data from both frames. Third, to implement DVIO on a smartphone with variable camera intrinsic parameters (CIP), a method called CIP-VMobile is devised to simultaneously estimate the intrinsic parameters and motion states of the camera. CIP-VMobile estimates in real time the CIP, which varies with the smartphone's pose due to the camera's optical image stabilization mechanism, resulting in more accurate device pose estimates. Various experiments are performed to validate the VI-SLAM methods with the two robotic assistive devices.

Beyond these primary objectives, SM-SLAM is proposed as a potential extension for the existing SLAM methods in dynamic environments. This forward-looking exploration is premised on the potential that incorporating dynamic object detection capabilities in the front-end could improve SLAM's overall accuracy and robustness. Various experiments have been conducted to validate the efficacy of this newly proposed method, using both public and self-collected datasets. The results obtained substantiate the viability of this innovation, leaving a deeper investigation for future work.


© The Author

Is Part Of

VCU University Archives

Is Part Of

VCU Theses and Dissertations

Date of Submission


Included in

Robotics Commons