Validation of Blaze-Pose as Model in the Analysis of Lower Extremities During Gait

The concept of healthcare is being transformed with the introduction of technologies such as computer vision. Recent advances have proven that computer vision implementations result as efficient and cost-effective solution for applications such as clinical gait analysis. Those, the objective of this research was to validate a computer vision-based human pose detection model named Blaze Pose, in its application for gait detection and angular analysis of the lower extremities in comparison with the manual parameter extraction using a biomechanical analysis software Kinovea. For this, the study emphasizes in walking task, specifically the joint angulations (hip, knee and ankle) observed from the lateral side. The results demonstrate that it is feasible to obtain joint angulations using the human pose estimation model. However, significant discrepancies in the results were found when comparing the ankle joint estimation, but there was a relevant similarity in the hip and knee joints; a 37% and 34% percentage error was also observed in the angular measurements obtained. This result indicates that improve in the model must be made to be able to apply it as a tool in clinical assessment o biomechanics.