Abstract

Kinesiologists who study people's posture during walking depend on spreadsheets and visual posture reviews. Gold-standard evaluation relies on expert evaluation, not mediated by technology. However, today there are technological advances to automate specific processes adequately. Our proposal focuses on developing software based on computer vision and artificial intelligence (AI) to support recognition in the gait cycle and walking activities. The software is deployed in an architecture based on microservices to support the image analysis process with high concurrency. We opted for an open-source alternative, Openpose, because it is one of the most popular detection libraries for pose estimation and is capable of real-time multi-person pose analysis. We validate the choice through a proof of concept in which we prove that it can be possible to obtain valuable results for the kinesiology care process. This software assists specialists in analyzing and measuring lower extremity angles and distances during gait. We developed an information system based on open-source pose estimation algorithms for clinical decision-making. The technological approach was obtained by analyzing similar proposals and considering the characteristics of the clinic. We used a real-time multi-person pose estimation as an essential element enabling machines to visually comprehend and analyze humans and their interactions. In this instance, we identified accuracy metrics and optimized the evaluation process time. Using a non-probabilistic sample, we analyzed the videos of users performing the gait exercises. These results indicate that although the algorithms still need to achieve perfect accuracy, they save manual work for the final evaluation. On average, using the platforms reduces by about 50% the total time required to generate the final reports delivered by the kinesiology clinic. This proposal has always been justified as a support to the professional work and not as a replacement. We propose an information system based on open-source pose estimation algorithms for clinical decision-making. The technological approach was obtained by analyzing similar proposals and considering the characteristics of the clinic. We used a real-time multi-person pose estimation as an essential element enabling machines to visually comprehend and analyze humans and their interactions. While these recognition alternatives have been explored for some time, linking with particular needs and improving healthcare processes is critical.