MRCS: multi-radii circular signature based feature descriptor for hand gesture recognition

Deaf and hearing-impaired persons communicate by means of signs and gestures. In course of time, this form of communication has evolved as natural languages with its own grammars and lexicons. Automatic hand gesture recognition is an important task in development of human computer interaction system for deaf mute community. In this paper, we report the development of a novel feature descriptor named Multi-Radii Circular Signature (MRCS) and associated automatic hand gesture recognition pipeline. This descriptor has certain desirable aspects such as translation, scale and rotation invariance, variable number of feature extraction, and symbol reconstruction. Multiple sets of experiments for various feature combinations with multiple classifiers have been carried out on three publicly available benchmark datasets viz. NTU 10-gesture dataset, HKU EEE DSP dataset and Senz3D dataset. Consistently high performance across multiple datasets and feature combinations reveals the robustness and generality of the descriptor. Its code and usage guidelines are also released at https://github.com/iilabau/MRCS for greater interest.