Development of glass and transparent obstacle detection algorithm for robotic wheelchair navigation

• Main Author: Goh, Hans
• Other Authors: Ang Wei Tech
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering; Robotics; Navigation; Transparent
• Online Access: https://hdl.handle.net/10356/177174
• Institution: Nanyang Technological University

Powered wheelchairs are difficult to control, especially for patients with mobility impairments. The need for robotic wheelchairs is eminent. Current navigation methods using LiDAR sensors are unable to detect glass obstacles. As wheelchairs frequently encounter glass obstacles like glass walls and doors, it could be dangerous for the patient if a robotic wheelchair is unable to detect these obstacles. From research on past work, there were several attempts with different methods adopted, such as using the intensity information of reflected laser from LiDAR or using glass frame information to predict the presence of glass. There were also attempts that leveraged sensor fusion, fusing the inputs of LiDAR as well as ultrasonic sensors, which could detect glass. Finally, methods that used computer vision techniques, were also developed in previous studies. After evaluating the pros and cons, as well as the feasibility of each approach, the computer vision route was chosen. Another factor that contributed to this decision was that the wheelchair being developed already had the required cameras installed and implemented for another function. Hence almost no extra work was required. The computer vision techniques developed from previous studies were first evaluated to see if they were feasible to be integrated onto the wheelchair. Then, they were compared, and the best approach was chosen. However, the best method was incomplete, and a novel interpolation algorithm was proposed. The developed method was integrated onto the RRIS Robotic Wheelchair and ran with a mapping algorithm to generate a map. The generated map was compared with a baseline map that was generated with LiDAR sensor. The results show that the proposed method successfully detected the presence of glass, and the glass was added onto the map.