Login | Register

3D human pose reconstruction for ergonomic posture analysis

Title:

3D human pose reconstruction for ergonomic posture analysis

Chu, Wenjing (2018) 3D human pose reconstruction for ergonomic posture analysis. Masters thesis, Concordia University.

[img]
Text (application/pdf)
Chu_MASc_S2019.pdf - Accepted Version
Restricted to Repository staff only until 18 January 2021.
Available under License Spectrum Terms of Access.
2MB

Abstract

The rapid development of the modular construction industry has produced the social concerns about workers’ health and safety in the factory-controlled construction processes. According to the reports from Association of Workers’ Compensation Boards of Canada (WCBC), approximately 2 in 100 workers are injured due to their awkward and improper postures and motions in the modular construction industry in Canada. The occurrence of injuries and accidents not only reduces the productivity but also increases the project cost. In this respect, the ergonomic posture built upon the self-report, manual observations, direct measurement or computer vision, is essential to identify, mitigate and prevent these postures for safety and productivity improvement. Advanced computer vision technologies have made the vision-based ergonomic posture analysis cost-effective in real workplaces. So far, several vision-based methods have been created to obtain the anthropometry data, such as joint coordinates and body angles, which are required for the ergonomic posture analysis. However, there are still some challenges like occlusions and lack of accuracy in complex working environments to reduce the reliability and robustness of these vision-based methods in practice.
This research proposes a novel framework that acquires the body joint angles for ergonomic posture analysis by reconstructing the 3D worker body with the 2D videos recorded from a monocular camera. The framework consists of (1) human tracking in the given videos; (2) 2D body joints and body parts detection using the tracking results; (3) 2D pose refining based on integrating the 2D joints detection with the body parts detection; (4) 3D body model generation and body angle calculation; and (5) ergonomic posture analysis based on the obtained body angles. The proposed framework has been tested on the videos in real factories and the test results were compared with the motion data captured by the IMU-based suit. The results showed that the average 3D pose difference was 17.51 degrees in terms of joint angles and the lowest joint angle difference was around 4 degrees.

Divisions:Concordia University > Gina Cody School of Engineering and Computer Science > Building, Civil and Environmental Engineering
Item Type:Thesis (Masters)
Authors:Chu, Wenjing
Institution:Concordia University
Degree Name:M.A. Sc.
Program:Civil Engineering
Date:October 2018
Thesis Supervisor(s):Zhu, Zhenhua and Han, Sang Hyeok
ID Code:984926
Deposited By: Wenjing Chu
Deposited On:17 Jun 2019 19:03
Last Modified:17 Jun 2019 19:03
All items in Spectrum are protected by copyright, with all rights reserved. The use of items is governed by Spectrum's terms of access.

Repository Staff Only: item control page

Downloads per month over past year

Research related to the current document (at the CORE website)
- Research related to the current document (at the CORE website)
Back to top Back to top