Development of a robotic guide dog

• Main Author: Ong, Albert Jia De
• Other Authors: Li King Ho Holden
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Mechanical engineering::Robots
• Online Access: https://hdl.handle.net/10356/167182; https://hdl.handle.net/10356/157751
• Institution: Nanyang Technological University

In this study, the robotic guide dog is further improved in the aspect of real-world application, where the robotic dog is able to detect pedestrian crossing traffic lights with relatively high accuracy for about 70% and guide the user across the road when the signals changes. For test 1, the performance of the model is tested using ZED camera to detect in lab setting a traffic light prop, where it performed relatively well within short distances of about 20cm away from the traffic light prop, with the average accuracy level over 30 runs of experimentation higher than 90% for both Pedestrian-Green and Pedestrian-Red. Further tests were done on 30 different first-person videos filmed by the author crossing roads while capturing the pedestrian crossing traffic light under different weather and light conditions. Initial results for Test 2 using Test 1 weights were not ideal as the percentages for all 3 parameters (accuracy, precision, and recall) were very low. However, after considerations were made to improve the training model through training with pedestrian crossing traffic light images taken during the day and night, and including images where the camera was 7m to 10m away from the pedestrian crossing lights on top of the images from the dataset, the results from Test 2 showed improvement across the board for the 3 parameters: accuracy of the model rose by 39.36%, precision by 38.78%, and recall by 46.97%. The outcome showed that the basic function of the robotic guide dog being able to detect pedestrian crossing traffic light in the real-world setting was able to work as intended. All in all, this proof-of-concept robotic guide dog shows promise in being able to be implemented in real world setting to assist the visually impaired community with their daily needs, especially for those that needs to travel about. This would also bridge the gap that some of the visually impaired users may face in terms of accessibility in certain areas, or places, allowing for the quality of lives of more visually impaired people to be improved.