Presentation
Evaluating the Accuracy of AI-Powered Ergonomic Assessments Using a Commercial Computer Vision System
DescriptionWorkers engaged in material handling tasks face a high risk of work-related musculoskeletal disorders (WMSDs). While recent AI-driven computer vision systems claim to assess ergonomic risk factors with minimal manual input, their accuracy remains largely unverified. We evaluated the accuracy of a commercial AI system in estimating key parameters of the Revised NIOSH Lifting Equation (RNLE) by comparing its outputs to those obtained from marker-based motion capture data (ground truth). Ten sex-balanced participants performed various lifting tasks, during which their movements were recorded using three cameras positioned at different angles. Simultaneously, 3D motion capture data were collected for comparison. The video recordings were uploaded to the commercial AI software, which analyzed the footage and extracted RNLE parameters. These outputs were then compared to the motion capture data to assess the AI system’s accuracy. Results revealed notable inaccuracies in the commercial system’s estimates, particularly for horizontal and vertical distances. These errors led to overestimated Recommended Weight Limits (RWL) and underestimated Lifting Index (LI) values. Among the three cameras, the side view yielded the most accurate results, while the moving camera produced the least reliable estimates. These findings suggest that current commercial AI-based ergonomic tools require substantial improvement before reliable workplace use.
Event Type
Lecture
TimeTuesday, October 14th12:10pm - 12:30pm CDT
LocationGrand C/D South
Occupational Ergonomics
Similar Presentations