Özet:

Abstract Hand gesture is an important non verbal communication mechanism of Indian classical dances especially Bharatanatyam. The hand gestures in Bharatanatyam are called as mudras and there are total 52 mudras with 28 single hand mudras and 24 double hand mudras. Many computer aid mudras classification systems were designed to infer the non verbal theme communicated via mudras. But unlike other hand gesture recognition system, accurate classification of mudra is challenging due to high structural similarity between mudras. This work proposesdeep learning multi scale feature guided aggregated signature for accurate classification of mudras. The deep learning multi scalefeatures are extracted from multi scale images after super resolution and thus self similarities between mudras can be easily differentiated. In addition the features are scale and orientation invariant. Aggregation signature is constructed based on multi scale super resolution features to reduce the classification time. The proposed solution is able to provide an average accuracy of 96% which is atleast 2% higher compared to existing works. Finally the proof of concept of application of proposed mudra classification system in augmented reality based learning system is presented. 

Anahtar Kelimeler: