An automatic ergonomic evaluation based on human movement recognition from motion capture data with random forest classifiers

Aim: This study aims to develop an automated system for classifying industrial operator movements using motion capture (MoCap) data and a random forest (RF) classifier, with the ultimate goal of supporting ergonomic risk assessment and identifying physically demanding tasks suitable for automation in human–robot collaborative environments. Methods: MoCap data were acquired using the Xsens system from nine participants (four men and five women) performing three movement classes: walking, standing, and bending. A labelled dataset was constructed and used to train an RF classifier via a sliding-window approach with statistical and frequency-domain feature extraction. The model was evaluated on a held-out test set (80/20 split) and through leave-one-subject-out cross-validation. The trained classifier was subsequently applied to a real industrial use case at a logistics company. MoCap data were additionally imported into IPS IMMA to generate a human digital twin and perform frame-by-frame Rapid Entire Body Assessment (REBA) trunk position scoring during identified bending movements. Results: The RF classifier achieved 98.44% accuracy on the test set, with F1-scores of 0.98, 0.97, and 1.00 for bending, standing, and walking, respectively. Leave-one-subject-out cross-validation consistently exceeded 91% accuracy across all participants. In the industrial use case, bending accounted for 6.7% of the total task time (127.5 s). During bending, the operator spent 45.98% of bending time in REBA trunk position zone 3 (20°–60° flexion) and 28.79% in zone 4 (> 60° flexion). Conclusions: The integration of MoCap data with RF-based movement classification and automated REBA trunk position extraction provides an objective, reproducible approach to ergonomic risk assessment in industrial settings. The results demonstrate that bending movements expose operators to elevated musculoskeletal risk, supporting the case for targeted automation of these tasks. This work constitutes a first step toward a broader framework for automated ergonomic evaluation and human–robot collaborative workstation design.