Novel Input of Continuous Recognition of Hand Gesture

EasyChair Preprint 7700

Abstract

This paper is to illustrate air-writing mechanism of different gesture by using deep learning techniques. Mainly, focus is given to observe and identify air-writing actions which perform unbreakable continuous gesture trajectory. In this paper, we attempt to eliminate unnecessary finger motions unrelated to literature or writing activities as compared to standard pattern recognition approach. For the experimental purpose, Xbox is used for finger tracking without using markers or gloves and also used datasets of writing and nonwriting finger actions.Window-based method for detecting and extracting the air-writing events from incessant watercourse of gesture data, which includes intended state unconnected in writing. A writing segment is created when multiple writing events occur in a row. Writing portion is used as a feed in deep learning network in recognizing different hand gesture trajectory, to create an exclusive method to develop air-writing system which includes both finding and acknowledgement stages. We propose dynamic hand gesture system which achieves an overall accuracy of 97% for word-based recognition and 94% for letter-based recognition using leave-one-out cross validation.

Keyphrases: AirWR, DHGR, Gesture Trajectory, HMM, finger-writing

@booklet{EasyChair:7700,
  author    = {Prashant Richhariya and Piyush Chauhan and Anita Soni and Sanjay Srivas},
  title     = {Novel Input of Continuous Recognition of Hand Gesture},
  howpublished = {EasyChair Preprint 7700},
  year      = {EasyChair, 2022}}