Bharatna̅~yam Adavu Recognition from Depth Data

Human motion recognition is a challenging problem having applications in multifaceted domain. It has significance in sports, surveillance, human-computer interaction etc. Proposed Bharatna̅ yam Adavu Recognition System from Depth Data (BARSDD) recognizes the Adavu in Bharatna̅ yam using extended state model. Bharatna̅ yam is one of the Indian traditional dance forms. A specific sequence of poses in Bharatna̅ yam is known as Adavu. Bharatna̅ yam dances are based on stories of Rama̅yana and Maha̅bha̅rata, the Indian Epics. Only, expert can understand this sentiment (bha̅va) and posture (mudra̅). Main objective behind this work is that BARSDD system should recognize meaning of dance and display it for audience. We have represented Adavu using extended state model. Five basic Adavu Tatta, Natta, Sarikal, Visharu, and Kuditta̅metta̅ are considered for recognition. Kinect sensor is used to achieve more reliable, accurate and cost-effective depth image sequence. Contour features extracted using Canny Edge Detection algorithm are used for representation of dancer's posture. Freeman Chain Code Algorithm followed by compression using run length encoding is applied to store contours. System is tested for all five Adavu performed five times each. Our system shows average 78.38% recognition rate. It is observed that failure of recognition is due to non-identification of some intermediate posture in Adavu.