Exploratory Ergonomic Risk Assessment Using Smart Insoles and Computer Vision

Traditional observation-based ergonomic assessments, like the REBA (Rapid Entire Body Assessment), remain effective tools for risk assessment in the workplace. Existing automated methods that use IMU sensors or cameras/computer vision techniques to capture human motion have facilitated data collection but may not be ideal for various applications. For example, IMU signals can be easily distorted in the presence of metallic objects or electromagnetic fields, and computer vision techniques require workers to remain in a relatively fixed area. An emerging non-obtrusive tool that allows workers to remain mobile is the use of pressure insoles.

In this exploratory study, we propose a novel approach that combines smart insoles with computer vision to assess ergonomic risk in a minimally invasive manner. Initially, we use computer vision algorithms to capture posture data, which provides the input needed to obtain REBA scores, serving as ground truth for training our model. Simultaneously, we collect plantar pressure data from smart insoles embedded with 16 sensors per foot.

Our methodology aims to map the pressure distribution data from insoles onto a risk of injury, potentially enabling risk assessment based solely on plantar pressure distribution. We will use 60% of the data for training our machine learning model, 20% for validation, and apply k-fold cross-validation to refine the model. The remaining 20% will be reserved for an independent assessment of the model’s performance. This project may contribute to developing a less intrusive, data-driven model for injury risk prediction, facilitating field-based risk assessments of injury.