Human Action Recognition and Coding based on Skeleton Data for Visually Impaired and Blind People Aid System

This work aims to contribute to the development of a novel device, perceivable by the touch sense, that allows visually impaired and blind persons understand the actions of the people around them. Therefore, we propose an end-to-end system that takes skeleton data, captured by kinect sensor, as input data and recognizes the performed human actions in the surrounding scene using a state-of-the-art GCN model (MS-G3D). To detect transitions between two actions in untrimmed video streams, a new machine learning (SVM) based method is integrated. It helps achieve less computational cost. Finally, we propose a novel human action coding to label the recognized action that will be mapped on the tactile device.