A Robust Silhouette-Based Human Action Recognition System Using Template Matching

Abstract

To identify human actions, this research introduces a novel method that uses silhouette and template matching. From silhouettes, the system identifies the action features, and a template generation method making use of silhouette extraction and averaging is proposed for recognizing the actions robustly. The experimental results have been performed on Weizmann dataset; experimental outcomes show that the proposed system achieves accuracy 95.8% better than the present methods. The scenarios where the proposed system works include walking, running, and jumping or even offering robustness from noise and fluctuations in data inputs.