Can a Deep-Learning Markerless Motion-Capture System Outperform the Gold Standard Marker-Based System? Concurrent Validity, Test-Reliability, and Usability

Abstract Marker-based optical motion capture systems have been used as a cardinal vehicle to probe and understand the underpinning mechanism of human posture and movement, but it is time-consuming (a few hours) for complex and delicate data acquisition and analysis, labor-intensive with highly trained operators. To mitigate such inherent issues, we developed an accurate and usable (5-minute data collection and processing) deep-learning-based 3-Dimensional markerless motion capture system called “Ergo,” designed for use in ecological digital healthcare environments. We investigated the concurrent validity and the test–retest reliability of the Ergo system measurement’s whole body joint kinematics (time series joint angles and peak joint angles) data by comparing it with a standard marker-based motion capture system recorded during an overhead squat movement. Our deep learning system analyzed 939,600 image data obtained from 71 healthy adults and successfully achieved optimal data processing (accurate, faster, and simpler). The Ergo system demonstrated excellent agreement for time series joint angles (r2= 0.88-0.99) and for peak joint angles (ICC2,1 = 0.75-1.0) when compared with the gold standard marker-based motion capture system. Additionally, we observed high test-retest reliability (ICC1,1 = 0.92-0.99). In conclusion, the deep learning-based markerless Ergo motion capture system considerably outperformed the Gold Standard marker-based motion capture system measurements in the concurrent accuracy, reliability, and usability (faster and simpler data collection and processing), thereby making it a highly accessible choice for diverse universal users and ecological industries or environments.