Author: Pan, Jun
Title: Real-time occlusion-aware human pose estimation for home scoliosis rehabilitation
Degree: M.Phil.
Year: 2022
Subject: Scoliosis -- Treatment
Telecommunication in medicine
Patient monitoring -- Data processing
Hong Kong Polytechnic University -- Dissertations
Department: Department of Computing
Pages: 74 pages : color illustrations
Language: English
Abstract: Computer science and technologies is changing our life by replacing labour in many areas. Telemedicine is also a popular direction to put machine learning in, where doctors can be replaced by artificial intelligence. Among the telemedicine capable diseases, scoliosis is a common disease in any age groups of human being. One popular rehabilitation approach to mild symptom scoliosis is exercise which can be tracked by human pose estimation approaches. Human pose estimation is an important research branch in computer vision. Despite of the benefit it takes to fitness, films and other services, there are still many challenges to be faced in such area, two of them are real-time on-device human pose estimation and occlusion-aware human pose estimation. This thesis aims to propose a novel occlusion handling method in human pose estimation to help better replace professional guider in scoliosis rehabilitation process, hence we are going to propose a general real-time on-device and occlusion-aware pipeline of human pose estimation which improved the performance both on accuracy and run-time. In the Chapter (§1), we introduce the motivation of combining human pose estimation and telemedicine to help rural patients rehabilitate from scoliosis and basic concepts of occlusion in human pose estimation. In the Chapter (§2) we present a literature review for better understanding the trend of occlusion handling and inference acceleration in human pose estimation and provide some perspective of solution. In the following two chapters, we provide two components in the system to solve the occlusion problem and enable the network of real-time ability. In the Chapter (§3), we propose a two-step semi-supervised learning pipeline to fully emphasize the anti-occlusion ability out of transformer-based autoencoder. In the Chapter (§4), we propose a general and task-independent Patch Automatic Skip Scheme (PASS), a novel end-to-end learning pipeline to support diverse video perception settings by decoupling acceleration and tasks. Both methods are general and task-irrelevant. Conclusions and suggestions will be provided in the Chapter (§5).
Rights: All rights reserved
Access: open access

Files in This Item:
File Description SizeFormat 
6669.pdfFor All Users17.22 MBAdobe PDFView/Open

Copyright Undertaking

As a bona fide Library user, I declare that:

  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
  3. I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.

By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.

Show full item record

Please use this identifier to cite or link to this item: