A Novel Approach for Gesture Spotting in an Assamese Gesture-Based Character Recognition System using a Unique Geometrical Feature Set

Abstract

Gesture Recognition (GR) systems have gained tremendous importance in recent times as these have established themselves as important elements of Human Computer Interaction (HCI). A significant attribute of GR systems is Gesture-Based Character Recognition (GBCR). GBCR systems serve as a powerful mediator for communication among people having hearing and speech impairments as well as with common people. They can also serve as a rehabilitative aid for people with motor disabilities who cannot write with pen on paper, or face difficulty in using common human-machine interactive (HMI) devices. A challenging task in GBCR system is gesture spotting, i.e. determining the start and end points of different segments in a character sequence. In this paper, we have addressed the design of an Assamese gesture-based character spotting system, which spots the different segments present in an overall character sequence. Partitioning of character segments is done by employing a unique geometrical feature set comprising of three features namely linear eccentricity, flatness and area of ellipse. The performance of our proposed system is validated by taking into account the vowels and numerals of Assamese alphabets. Moreover, the efficacy of our proposed hand segmentation module enables the system to tackle different background conditions like complex background containing multiple objects and scenes involving multiple signers in the background.