Accuracy of An Algorithm Using Motion Data Of Five Wearable IMU Sensors For Estimating Lifting Duration And Lifting Risk Factors

Abstract

The objective of this study was to assess the accuracy of an algorithm for processing data from five inertial measurement unit (IMU) sensors for measuring the vertical distance (V) and horizontal distance (H) of two handed lifting, trunk flexion angle (T) and lifting duration (LD). The sensors were placed on five body segments including the left wrist, right wrist, upper arm of the dominant hand, upper back, and thigh of the dominant leg. A laboratory-grade optical motion capture system was used as the ground truth for the assessment. Data were collected on ten subjects that performed 12 two-handed lifting tasks varying in height of the hands and horizontal distance between the body and the lifted object. Results showed that the algorithm performed well for determining the LD (~1 sec error) and T (~2° error). The average errors for V and H were about 33 and 6.5 cm, respectively.

A deep learning approach for lower back-pain risk prediction during manual lifting

“…Data collection of the set used to train the proposed model is further described in [ 13 ]. Notably, the object lifted—an 0.45 kg wire grid—was designed to minimize the impact of any possible injury from lifting incorrectly.…”

“…For purposes of analysis, data utilized for model development was sourced from a previous study by researchers at NIOSH to examine body motion for two-handed lifting tasks similar to those in the workplace [ 13 ]. Lifts were performed using the American Conference of Governmental Industrial Hygienists (ACGIH) Threshold Limit Values (TLV) for lifting, which defines 12 zones relative to the body in the sagittal plane shown in Fig 1 .…”

“…Each sensor consisted of a tri-axial gyroscope and accelerometer sampling at a rate of 25 Hz. All sensor data was calibrated and synchronized prior to data collection [ 13 ]. An example of this data is shown in Fig 2 .…”

“…Additionally, sourcing data for specialized activities has its own challenges, given a dearth of existing datasets and the increased expense of independently collecting data. Consequently, datasets are typically small; the NIOSH lifting dataset consists of 720 total trials [ 13 ]. The challenge, then, is to develop a versatile model that can distinguish between very similar activities and operates on datasets possibly magnitudes smaller than for similar problems [ 14 ].…”

Characterizing Human Box-Lifting Behavior Using Wearable Inertial Motion Sensors

“…These orientations were used to calculate the trunk flexion angle, the vertical height of the object, and the horizontal distance from the object to the lifter. In a second related study [25], the authors compared the estimated values to values measured by a commercial motion capture system and found that estimation of the vertical and horizontal positions of the box were poor with mean errors of 33 cm and 6.5 cm, respectively. The estimate of trunk flexion had a mean error of 2.3 degrees.…”