| The primary aim of this study was to develop machine learning or deep-learning aided
procedures along with scientific investigations that enhances the capability of a commercial non-
autonomous wheelchair towards autonomy. The thesis addresses the computer vision work
for obstacle detection and localization applied to an autonomous wheelchair operation. The
computer vision tasks including the depth image classification are accommodated in a small
form factored and resource constraint computers such as Raspberry Pie and Google Coral. The
tasks and strategies also include classifying the images using a pretrained model (TensorFlow
lite), detecting and measure the degree of obstacle avoidance by pairing color (RGB) image
classification with depth images. The thesis also offers approaches for indoor localization
applicable for the autonomous wheelchair development. The objective has been further extended
to develop a simulation platform for autonomous wheelchair driving where navigation and path
mapping construction algorithm evaluations are visually offered using MATLAB®. In addition,
the thesis includes research and project contributions prior to the change of thesis subject to
autonomous wheelchair development. These contributions are addressed as the additional works
which includes (1) the initial work on error determination in motion capture process using
VICON and (2) a wafer alignment fault detection process using image processing.
Keywords: machine learning, computer vision, autonomous wheel-chair, navigation, mobile robot,
depth image, obstacle avoidance. |