Automatic real-time occupational posture evaluation and select corresponding ergonomic assessments

Abstract

The objective is to develop a system to automatically select the corresponding assessment scales and calculate the score of the risk based on the joint angle information obtained from the imaged process (OpenPose) via image-based motion capture technology. Current occupational assessments, for example, REBA, RULA, and OWAS were used to evaluate the risk of musculoskeletal disorders. However, the assessment result would not be reported immediately. Introducing real-time occupational assessments in different working environments will be helpful for occupational injury prevention. In this study, the decision tree was developed to select the most appropriate assessment method according to the joint angles derived by OpenPose image process. Fifteen operation videos were tested and these videos can be classified into six types including maintenance, handling, assembly, cleaning, office work, and driving. The selected ergonomic assessment method by our developed decision tree in each condition are consistent with the recommendation of the Labour Research Institute. Moreover, the high-risk posture could be identified immediately and provide to the inspector for further evaluation on this posture rather than the whole operation period. This approach provides a quick inspection of the operation movements to prevent musculoskeletal injuries and enhances the application of the scale assessment method in different industrial environments.

Evaluation of upper body postural assessment of forklift driving using a single depth camera

“…The CV system was robust to dynamic light conditions and the mean RMSE of all joint angles across different tasks is in line with earlier studies (Kim et al, 2021;Lin et al, 2022). Figure 3 shows that in comparison with other joint angles, the head flexion/extension angle had higher variance across tasks.…”

“…Based on the findings discussed above, this study was unable to confirm the applicability of the proposed CV system directly for RAMP ergonomic assessment in industry, even though previous studies from lab environments suggested otherwise (Kim et al, 2021;Lin et al, 2022;Massiris Fernández et al, 2020).…”

“…In line with that, low-cost MoCap systems based on computer vision (CV) have been researched to solve the aforementioned posture assessment problems. One such MoCap solution is OpenPose, a realtime multi-person 2D pose estimation system, which has been used for postural assessments with both 2D (Lin et al, 2022) and 3D approaches (Kim et al, 2021). These studies have shown that OpenPose has acceptable accuracy for posture assessment with joint angle errors of 8.0° and 8.4° respectively.…”

Automated Detection of Improper Sitting Postures in Computer Users Based on Motion Capture Sensors

“…Image processing algorithms are often combined with the application of neural networks [17,19,20], which results in high classification accuracy for both person-specific and general classification tasks. Specifically, the system presented in Ref [18] has a reported recognition rate of high-risk work postures comparable to the baseline assessment of an ergonomic expert. Overall, image-processing-based approaches provide a non-disruptive and accurate approach for classifying work postures, which do not require wearable sensors or devices and are most commonly employed for manufacturing or handling tasks.…”

“…These solutions employ camera and imaging setups that record workers and access their work posture and the connected health risks. This approach can be used for computer-related office work [17] and physically demanding high-risk work, such as maintenance, handling and assembly [18]. Image processing algorithms are often combined with the application of neural networks [17,19,20], which results in high classification accuracy for both person-specific and general classification tasks.…”