Please use this identifier to cite or link to this item: https://hdl.handle.net/10356/154661
Title: Human-aware robot navigation for assistive wheelchair
Authors: Yeoh, Yong Shan
Keywords: Engineering::Mechanical engineering::Robots
Issue Date: 2021
Publisher: Nanyang Technological University
Source: Yeoh, Y. S. (2021). Human-aware robot navigation for assistive wheelchair. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/154661
Abstract: Human tracking is an important task in a robotic wheelchair operating in a human crowded environment. It enables the robot to interact with humans in the environment effectively while ensuring their safety. However, this is a challenging task as human crowded environment are unstructured in nature. Unexpected events such as occlusions or non-linear human motion occurs frequently. The task becomes more difficult when the camera is mounted on a mobile robot platform. Camera motion results in a change in perspective of the camera, and thus a change in appearance of the targets. Furthermore, if camera motion is unaccounted for in the motion model of the human tracking system, an unexpected motion of the targets will be perceived by the system. In this report, we present a Multiple Object Tracking (MOT) module towards achieving realtime human tracking on a robotic wheelchair platform. The MOT module utilizes a YOLOv4 model for object detection. Camera projection matrix is used to convert the bounding box to a point in the world coordinates system. A Kalman Filter is used to track motion of the targets. A ReID network is used to generate an appearance descriptor of the detections. Track association is performed by evaluating a combined metric that describes both the positional distance and appearance distance of detections in the current time step compared to those in previous time step. Apart from a human tracking system, this report will also present the design and implementation of a tracking camera and a human following module. These implementations complement the MOT module in enhancing the capabilities of the robotic wheelchair. Experimental results validate the performance of the proposed MOT module to track humans in a real-world environment. The proposed MOT system managed to outperform existing MOT approaches such as Deep SORT in terms of accuracy without a compromise in speed.
URI: https://hdl.handle.net/10356/154661
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:MAE Student Reports (FYP/IA/PA/PI)

Files in This Item:
File Description SizeFormat 
FYP Final Report_Yeoh Yong Shan_Amended.pdf
  Restricted Access
2.04 MBAdobe PDFView/Open

Page view(s)

80
Updated on May 17, 2022

Download(s)

26
Updated on May 17, 2022

Google ScholarTM

Check

Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.