Skeleton Based Micro-Gestures Classification
Renzhi Wang, JianHua Hu, Zhanjiang Yuan · 2024
Micro-gestures are small., subtle movements made by individuals that convey important information about their emotions., thoughts., and intentions. However., these movements are often difficult to detect and interpret by humans., always., they are ignored. Micro gestures are often unintentional behaviors driven by human's inner feelings., different from gestures such as sign language which performed for illustrative purposes. One significant challenge in this field is the interpretation of micro-gestures., the interpretation of these gestures remains a difficult task due to their ambiguity and context-dependence. Therefore., developing robust and context-aware algorithms that can accurately interpret micro-gestures is an interesting and important research. This research discusses the challenges associated with recognizing micro-gestures using skeleton-based methods., including feature extraction and classification. There are several existing approaches to analyzing and classifying micro-gestures., including handcrafted feature extraction techniques and deep learning architectures. We provide a Micro-gesture analysis system combining handcrafted feature and deep learning model to classify the Micro-gesture. We estimate our algorithm in two datasets (SMG datasets and iMiGUE datasets)., the result shows our method improves top-1 accuracy by 8 percentage compared with original ST-GCN.