Massiris Fernández, M., Bajo, J. M., Fernández, J. Á., & Delrieux, C. A. Joint Angle Estimation with VIBE: an Evaluation Using Virtual Avatars.

MANLIO MASSIRIS FERNÁNDEZ; JUAN M. BAJO; J. ÁLVARO FERNÁNDEZ; CLAUDIO A. DELRIEUX

Conferencia; conference IEEE ARGENCON 2020; 2020

Work-related musculoskeletal disorders (WMSD) are health problems of the human locomotion system caused or intensified by inadequate work activities or environments. The usual alternative to prevent WMSDs is to implement an ergonomics program, which must include an evidence-based ergonomic risk assessment to recognize and rectify ergonomic deficiencies in a wide range of workplace situations. Ergonomic risk assessment is done traditionally through the completion of specific forms, through in-site observation by ergonomists. This is inaccurate and inefficient due to subjective biases. Also, it is not practical due to the cost, time, and technical knowledge required. For this reason, in this work we propose the standardization of data collection through computer vision, which offers the opportunity to obtain a considerable replication level, increasing the reliability of the results and the quality of the data available to ergonomists. As a methodology, we propose using the open-source neural network VIBE to detect skeletal joints in photos whose purpose is to calculate ergonomic risk. As an approach to rich and reliable data gathering, we carried out a virtual experiment, using controlled avatars in the Unity game engine. This allowed us to explore and quantify acquisition performances, in particular the occlusion effects generated by varying the camera angles concerning the worker?s bodies, and the errors induced by background clutter.