Ergonomic Assessment with a Convolutional Neural Network. A Case Study with OWAS

The ergonomic assessment of workplaces is key to optimize work stations and tasks, and in some legal frameworks it is mandatory for companies to perform such evaluations. Currently, most of the evaluations are carried out by means of observational methods like REBA, OWAS, RULA, etc. The usual procedure to apply those methods consists of recording the worker in a normal work day, and then analyze postures, joint angles and movements by visual inspection. That is a tedious and slow process, which is very dependent on the evaluator’s experience. Artificial vision can help to analyze and objectify the video in a few minutes. The objective of this work is the development of an ergonomic assessment method based on artificial vision and a convolutional neural networks, in order to reduce the time of analysis of the videos used in ergonomic evaluations. For this purpose, the neural network used (Simple Pose network) and its application to the analysis of postures is presented, and then a use case is presented, evaluating a workplace with the OWAS methodology.