Toward a Fully Automated NIOSH Lifting Equation Using Computer Vision

Abstract

Objective

To develop a computer vision-based method for automating the Revised NIOSH Lifting Equation (RNLE) by identifying lift start and end times and estimating key RNLE multipliers from video data.

Background

The RNLE is widely used to assess low back injury risk during lifting tasks; however, its traditional application relies on manual measurements that are labor-intensive and prone to human error. Recent advances in computer vision offer opportunities to automate this process.

Method

The proposed method follows a three-stage process: (1) BlazePose, a real-time pose estimation model, detects 22 key body joints, followed by low-pass filtering to reduce noise. (2) Kinematic features are extracted from these joints, and video frames are classified as lifting or nonlifting to identify lifting phases and lift start and end times. (3) RNLE multipliers are estimated at these lift timings to compute the recommended weight limit (RWL).

Results

The method achieved a mean absolute timing error of 0.25 s for lift timing identification and showed strong correlations between estimated and ground-truth multipliers. The mean absolute error for RWL estimation was 1.58 kg, with a correlation coefficient of 0.91.

Conclusion

These results demonstrate the feasibility of using computer vision to automate the RNLE. Further improvements in lift timing identification, depth estimation, and validation in more diverse workplace settings are recommended.

Application

The proposed method enables RNLE implementation on widely available platforms such as mobile devices and surveillance cameras, promoting safer lifting practices in real-world workplace environments.