Presentation
Designing and Building a 3D Human Motion Dataset for Vision-Based Ergonomics Risk Assessments
DescriptionArtificial Intelligence (AI) is increasingly used in ergonomics, particularly for assessing musculoskeletal disorder (MSD) risks using computer vision-based evaluation. Recent advancements in vision-based AI have made it possible to monitor MSD risks using an ordinary camera, offering a more accessible and less intrusive alternative to traditional observation methods. However, existing AI models, often trained on generic datasets from the computer science domain, lack the keypoints necessary for calculating the intricate angles, especially in high-degree-of-freedom (DoF) joints, like distinguishing neck flexion/extension, lateral flexion, and rotation, for ergonomics risk assessments. We present the design and building process of a large-scale 3D human motion dataset designed to train vision-based AI models for ergonomics risk assessments. The dataset includes 47 custom-selected keypoints optimized for high-DoF joint angle calculations and visual features, capturing 7 million frames of 10 subjects performing 9 categories of manual material handling tasks under varied task parameters. A baseline MotionBert model trained on our dataset achieved a mean absolute angle error of 3.5° on the validation set and demonstrated generalization capability on real-world industry videos. Our work offers guidelines for collecting custom datasets and training AI models to enable more accessible, less intrusive monitoring of MSD risks using ordinary cameras.
Event Type
Lecture
TimeTuesday, October 14th1:50pm - 2:10pm CDT
LocationGrand C/D South
Occupational Ergonomics