Reliability of distal upper extremity posture matching using slow-motion and frame-by-frame video methods

Predicting Wrist Posture during Occupational Tasks Using Inertial Sensors and Convolutional Neural Networks

Current methods for ergonomic assessment often use video-analysis to estimate wrist postures during occupational tasks. Wearable sensing and machine learning have the potential to automate this tedious task, and in doing so greatly extend the amount of data available to clinicians and researchers. A method of predicting wrist posture from inertial measurement units placed on the wrist and hand via a deep convolutional neural network has been developed. This study has quantified the accuracy and reliability of the postures predicted by this system relative to the gold standard of optoelectronic motion capture. Ten participants performed 3 different simulated occupational tasks on 2 occasions while wearing inertial measurement units on the hand and wrist. Data from the occupational task recordings were used to train a convolutional neural network classifier to estimate wrist posture in flexion/extension, and radial/ulnar deviation. The model was trained and tested in a leave-one-out cross validation format. Agreement between the proposed system and optoelectronic motion capture was 65% with κ = 0.41 in flexion/extension and 60% with κ = 0.48 in radial/ulnar deviation. The proposed system can predict wrist posture in flexion/extension and radial/ulnar deviation with accuracy and reliability congruent with published values for human estimators. This system can estimate wrist posture during occupational tasks in a small fraction of the time it takes a human to perform the same task. This offers opportunity to expand the capabilities of practitioners by eliminating the tedium of manual postural assessment.

Assessment of Joint Angle and Reach Envelope Demands Using a Video-Based Physical Demands Description Tool

BACKGROUND

Current methods for describing physical work demands often lack detail and format standardization, require technical training and expertise, and are time-consuming to complete. A video-based physical demands description (PDD) tool may improve time and accuracy concerns associated with current methods.

METHODS

Ten simulated occupational tasks were synchronously recorded using a motion capture system and digital video. The tasks included a variety of industrial tasks from lifting to drilling to overhead upper extremity tasks of different cycle times. The digital video was processed with a novel video-based assessment tool to produce 3D joint trajectories (PDAi), and joint angle and reach envelope measures were calculated and compared between both data sources.

RESULTS

Root mean squared error between video-based and motion capture posture estimated ranged from 89.0 mm to 118.6 mm for hand height and reach distance measures, and from 13.5° to 21.6° for trunk, shoulder, and elbow angle metrics. Continuous data were reduced to time-weighted bins, and video-based posture estimates showed 75% overall agreement and quadratic-weight Cohen's kappa scores ranging from 0.29 to 1.0 compared to motion capture data across all posture metrics.

CONCLUSION AND APPLICATION

The substantial level of agreement between time-weighted bins for video-based and motion capture measures suggest that video-based job task assessment may be a viable approach to improve accuracy and standardization of field physical demands descriptions and minimize error in joint posture and reach envelope estimates compared to traditional pen-and-paper methods.

Wrist Posture Estimation Differences and Reliability Between Video Analysis and Electrogoniometer Methods

OBJECTIVE

The aim of this study was to determine the inter- and intrarater agreement of estimated wrist angles using video and to compare wrist angles from video analysis to electrogoniometers.

BACKGROUND

Video analysis is used frequently in ergonomic assessments, but factors including parallax and complex angles may influence wrist angle estimates. Electrogoniometers are an alternative to video, but may not be reliable in complex postures. Given the limitations of each method, there is a need to determine the suitability of the measurement methods for field use.

METHOD

Ten participants performed frame-by-frame wrist (flexion-extension, radioulnar deviation) and forearm (pronation-supination) posture estimation for worker tasks from three camera views (top, side, and oblique). Workers were equipped with electrogoniometers to record wrist posture during the tasks. The video estimate data was compared between 2 days and to sensor data.

RESULTS

Percent agreement between participants ranged from 53% to 81% across all ratings. Agreement was highest from the side view (66%, κ = 0.56) for flexion-extension and top view for radioulnar deviation (77%, κ = 0.52) and pronation-supination (69%, κ = 0.58). Video-electrogoniometer agreement was lower, with peak agreement from the top view for flexion-extension (57%, κ = 0.49) and radioulnar deviation (68%, κ = 0.30) and the oblique view for pronation-supination (53%, κ = -0.1).

CONCLUSION

Participant estimate agreement was moderate-substantial overall and aligns with previous reports. Disagreement between video and electrogoniometers may be attributed to camera angle and parallax effects and the small magnitude of wrist motions compared to other joints.

Accuracy of a Low-Cost 3D-Printed Wearable Goniometer for Measuring Wrist Motion

Wrist motion provides an important metric for disease monitoring and occupational risk assessment. The collection of wrist kinematics in occupational or other real-world environments could augment traditional observational or video-analysis based assessment. We have developed a low-cost 3D printed wearable device, capable of being produced on consumer grade desktop 3D printers. Here we present a preliminary validation of the device against a gold standard optical motion capture system. Data were collected from 10 participants performing a static angle matching task while seated at a desk. The wearable device output was significantly correlated with the optical motion capture system yielding a coefficient of determination (R2) of 0.991 and 0.972 for flexion/extension (FE) and radial/ulnar deviation (RUD) respectively (p < 0.0001). Error was similarly low with a root mean squared error of 4.9° (FE) and 3.9° (RUD). Agreement between the two systems was quantified using Bland–Altman analysis, with bias and 95% limits of agreement of 3.1° ± 7.4° and −0.16° ± 7.7° for FE and RUD, respectively. These results compare favourably with current methods for occupational assessment, suggesting strong potential for field implementation.

Inter-rater reliability for movement pattern analysis (MPA): measuring patterning of behaviors versus discrete behavior counts as indicators of decision-making style

The unique yield of collecting observational data on human movement has received increasing attention in a number of domains, including the study of decision-making style. As such, interest has grown in the nuances of core methodological issues, including the best ways of assessing inter-rater reliability. In this paper we focus on one key topic – the distinction between establishing reliability for the patterning of behaviors as opposed to the computation of raw counts – and suggest that reliability for each be compared empirically rather than determined a priori. We illustrate by assessing inter-rater reliability for key outcome measures derived from movement pattern analysis (MPA), an observational methodology that records body movements as indicators of decision-making style with demonstrated predictive validity. While reliability ranged from moderate to good for raw counts of behaviors reflecting each of two Overall Factors generated within MPA (Assertion and Perspective), inter-rater reliability for patterning (proportional indicators of each factor) was significantly higher and excellent (ICC = 0.89). Furthermore, patterning, as compared to raw counts, provided better prediction of observable decision-making process assessed in the laboratory. These analyses support the utility of using an empirical approach to inform the consideration of measuring patterning versus discrete behavioral counts of behaviors when determining inter-rater reliability of observable behavior. They also speak to the substantial reliability that may be achieved via application of theoretically grounded observational systems such as MPA that reveal thinking and action motivations via visible movement patterns.

Inter-rater reliability of a video-analysis method measuring low-back load in a field situation

Valid and reliable low-back load assessment tools that can be used in field situations are needed for epidemiologic studies and for ergonomic practice. The aim of this study was to assess the inter-rater reliability of a low-back load video-analysis method in a field setting. Five raters analyzed 50 work site manual material handling tasks of 14 workers. Peak and mean moments at the level of L5S1, and segment angles were obtained using the video-analysis method. Intra-class correlation coefficients (ICCs) and median standard deviations across raters were calculated. ICCs revealed excellent inter-rater reliability (>0.9) for peak and mean moments, ICCs of segment angles were variable. Median standard deviations showed relatively small inter-rater variance for moments (standard deviation <10 Nm) and segment angle variation ranging from 0° to 20°. The proposed video-analysis method, provides a reliable tool for obtaining low-back loads from occupational field tasks.

Work-site musculoskeletal pain risk estimates by trained observers--a prospective cohort study

Work-related musculoskeletal pain (MSP) risk assessments by trained observers are often used in ergonomic practice; however, the validity may be questionable. We investigated the predictive value of work-site MSP risk estimates in a prospective cohort study of 1745 workers. Trained observers estimated the risk of MSP (neck, shoulder or low-back pain) using a three-point scale (high, moderate and low risk) after observing a video of randomly selected workers representing a task group. Associations of the estimated risk of pain and reported pain during a three-year follow-up were assessed using logistic regression. Estimated risk of neck and shoulder pain did (odds ratio, OR: 1.45 (95% confidence interval, CI: 1.01-2.08); 1.64 (95% CI: 1.05-2.55)), however, estimated risk of low-back pain did not significantly predict pain (OR: 1.27 (95% CI: 0.91-1.79)). The results show that observers were able to estimate the risk of shoulder and neck pain, whereas they found it difficult to estimate the risk of low-back pain. Practitioner Summary: Work-related musculoskeletal pain risk assessments by observers are often used in ergonomic practice. We showed that observers were able to estimate shoulder and neck pain risk, but had difficulties to estimate the risk of low-back pain. Therefore, observers' risk estimates might provide a useful method for musculoskeletal pain risk assessments.

The validity and interrater reliability of video-based posture observation during asymmetric lifting tasks

OBJECTIVE

The objective was to evaluate the validity and interrater reliability of a video-based posture observation method for the major body segment angles during asymmetric lifting tasks.

BACKGROUND

Observational methods have been widely used as an awkward-posture assessment tool for ergonomics studies. Previous research proposed a video-based posture observation method with estimation of major segment angles during lifting tasks. However, it was limited to symmetric lifting tasks. The current study extended this method to asymmetric lifting tasks and investigated the validity and the interrater reliability.

METHOD

Various asymmetric lifting tasks were performed in a laboratory while a side-view video camera recorded the lift, and the body segment angles were measured directly by a motion tracking system. For this study, 10 raters estimated seven major segment angles using a customized program that played back the video recording, thus allowing users to enter segment angles. The validity of estimated segment angles was evaluated in relation to measured segment angles. Interrater reliability was assessed among the raters.

RESULTS

For all the segment angles except trunk lateral bending, the estimated segment angles were strongly correlated with the measured segment angles (r > .8), and the intraclass correlation coefficient was greater than 0.75.

CONCLUSION

The proposed observational method was able to provide a robust estimation of major segment angles for asymmetric lifting tasks based on side-view video clips. The estimated segment angles were consistent among raters.

APPLICATION

This method can be used for assessing posture during asymmetric lifting tasks. It also supports developing a video-based rapid joint loading estimation method.