Skeleton-Based Continuous Gesture Recognition Using Gesture Detection and Classification

Gesture is a common mode of communication in daily life. When integrated with human-computer interaction, it enhances convenience. However, continuous gesture recognition presents challenges due to potential ambiguity introduced by action-to-action continuity. This paper introduces a method for continuous gesture recognition using hand skeleton data. We propose two lightweight 1D-CNN s for detecting and classifying independent gestures. Leveraging single activation and data filtering concepts, we develop a system for continuous recognition. Our method, evaluated on the IPN dataset, achieves 81.28% accuracy with an inference speed of 4.75ms for isolated gesture recognition, and 54.63% accuracy with an inference speed of 20.8ms for continuous gestures recognition. Experimental results underscore the method's lightweight design and high accuracy.