Date of Award

7-8-2024

Publication Type

Thesis

Degree Name

M.A.Sc.

Department

Industrial and Manufacturing Systems Engineering

Keywords

Bio-signals;Computer vision;Facial analysis;Lifting risk assessment;NIOSH lifting equation;Physical workload classification

Supervisor

Eunsik Kim

Abstract

Lifting tasks pose significant risks to workers, necessitating effective risk assessment methods. The Revised NIOSH Lifting Equation (RNLE) helps evaluate these risks. RNLE manual assessments often introduce errors, while instrument-based methods can be costly, intrusive, and require technical staff to interpret the data. This study explores non-intrusive, cost-effective methods using computer vision for facial analysis and bio-signal sensors for physiological data to automatically classify lifting task risks as safe or risky. Ten lifting tasks were designed, collecting participants' facial data and bio-signals with a head-mounted camera and ECG and EDA sensors, labeled as safe or risky based on the RNLE. Three datasets (facial-only, bio-signals-only, and mixed) are analyzed to identify the most effective data type. The study also evaluates two models, Functional Data Analysis (FDA) and Recurrent Neural Networks (RNN), while using raw data to preserve information integrity, to find the optimal model for risk classification. Results showed that the RNN model trained on the mixed dataset achieved the highest mean accuracy (70%). This study demonstrates that lifting task risks can be estimated automatically using non-intrusive and cost-effective facial recordings and bio-signal sensors, offering an effective alternative for ergonomic assessment.

Download

Share

COinS