Human Action Recognition Based on Spatio-Temporal Feature Extraction from Skeleton Data

Compared with RGB data, human action recognition based on skeleton data does not contribute to the influences of background occlusion, changes of light, and viewpoints. Moreover, the quantity of data needed for the human skeleton data is small, they can increase the recognition speed and efficiency. This paper raises the problem left by conventional learning methods, that is, how to extract multi-scale temporal and spatial features from data samples insufficiently Due to this, the spatio-temporal feature extraction algorithm introduced in this paper that can extract both the spatial attitude features and the multi-scale temporal displacement features at the same time. This proposed algorithm is tested on two public sets and contrasted with other existing algorithms. These experiments reveal that the recognition performance of the proposed algorithm is better than the conventional algorithms.