Supervisors info:
Ηλίας Τσέπης, Καθηγητής, Τμήμα Φυσικοθεραπείας, Πανεπιστήμιο Πατρών
Όλγα Σαββίδου, Καθηγήτρια, Ιατρική Σχολή, ΕΚΠΑ
Βασίλειος Νικολάου, Καθηγητής, Ιατρική Σχολή, ΕΚΠΑ
Summary:
In the present work, a comparison was made between two motion capture systems. The optical motion capture system by VICON and the artificial intelligence program Pose2Sim (Pagnon et al., 2021) using five smartphones were evaluated.
For the purposes of this study, two different movements (deep squat and stair ascent) were recorded from four healthy individuals with the VICON cameras and the five smartphones simultaneously. The comparison of the two systems was based on the analysis of pelvic tilt, hip flexion, and knee angle. These data were chosen for their safety in observation, considering the selected movements and the optimal simulation of lower limb motions provided by Pose2Sim. We hope to explore upper limb movement results in future research by further adapting Pose2Sim to our needs. Pose2Sim allows users to detect two-dimensional coordinates from multiple videos using OpenPose (Cao et al., 2019). The architecture of OpenPose enables the extraction of limb coordinates from videos or photos, typically using the head or neck of the person as a reference point. This makes it possible to extract data from various video formats (webcams, smartphones, CCTV cameras, etc.), and its lightweight version allows usage even on low-processing power computers.
Additionally, Pose2Sim provides tools for camera calibration, object tracking in all videos, creation of «.trc» folders with three-dimensional coordinates, and filtering of results to eliminate noise in the data. The choice of Pose2Sim was based on its accessibility, open-source nature that provides flexibility according to researchers' needs, and its integration with the biomechanical simulation program OpenSim (Delp et al., 2007).
OpenSim is a widely used open-source program that calculates 3D joint angles using coordinate data from markers. In our research, it was chosen as a platform for comparing the results of Pose2Sim and the VICON system. The accuracy of OpenSim simulations, combined with the high-quality data obtained from VICON, allowed us to have a reliable measure for comparing the deviations between Pose2Sim results and the desired outcomes.
Keywords:
Artificial intelligence, Motion capture, Smartphones cameras, Computer vision, Opensource, Simulation
