Human Action Recognition Using Explainable Features and Sparse Motion History Images

This study proposes a novel approach to human action recognition (HAR) via depth sensor data.We introduce explainable features derived from skeleton sequences and a sparse motion history image (SMHI) to effectively capture motion characteristics.Our method addresses the limitations of current approaches by reducing the computational requirements while maintaining high accuracy.We propose a SlowFast-like network that combines these features for efficient HAR.Experiments on three datasets demonstrate the effectiveness of our approach, which achieves competitive accuracy with fewer features.The method also ensures user privacy by relying solely on skeleton data.This research contributes to the theoretical advancement of HAR and its practical application in various fields.