· Contributors · Organizations ·
Computer Vision Predicts Force During Lifting
Virtual Program Session
DescriptionHeavy lifting tasks are commonly observed and are a major cause of workplace injuries. The level of force exerted by workers remains challenging for practitioners to estimate. We propose a computer vision method for estimating injury risks due to varying force levels in lifting and demonstrate this method with common lifting tasks. The method utilizes computer vision techniques to extract features from workers' body motions, posture, and facial expressions. The extracted features were normalized and used to estimate the lifting risk levels determined by the NIOSH Lifting Equation. Specifically, two 1DCNN+LSTM classification models were developed; one of them was a binary classification model (safe or unsafe), and the other one was a three-level classification model (low-, medium-, or high-risk level). The binary and three-level classification model achieved an AUC and an accuracy of 0.890 and 0.721, respectively. In summary, this study proposes a novel non-intrusive method for lifting risk assessment.