0

Richa Golash

Research Associate in Bennett University, India

Title: Cognitive recognition and Interpretation of Dynamic hand Gestures in Webcam videos.

Abstract

Moving object detection, tracking, and converting its path into meaningful machine instruction, has become interesting field of research. But current state of art in the case of dynamic hand gesture recognition (DHGR), working with RGB images is facing many challenges due to natural and environmental factors associated with hand complex skeleton structure and unpredictable behaviour movement. Variations in 2-D coverage area of  hand postures, orientation and scale that occur due to behavioral characteristic of the subject (user) and physical factors like the real-time background, illumination set-up, camera view, etc. contribute to the loss of information when hand movement is captured through simple RGB cameras. 

A real-time DHGR system basically requires an ‘online’ detection strategy and an environment ‘adaptive’ tracking procedure. In general, to minimize these complexities advanced sensor-based or multiple cameras are preferred over simple RGB camera. To overcome the limitations of RGB videos, we have proposed a hybrid framework (comprising Faster-Region-based Convolutional Neural Network with hand-crafted feature extraction algorithm that is Scale-Invariant feature transform that ‘online’ detect the hand posture and ‘adapt’ the common  variations in the hand posture while performing real-time hand movement, independent of subject’s hand shape and size. 

We have also proposed a user-friendly Natural User Interface (NUI) via DHGR (integrating two DNNs) that is especially designed for seniors who are not capable to perform stable hand movement due to poor reflexes.
Our proposed system exhibits an efficiency of 98.76 % for hand detection and 95.83 % for tracking in real-time complex backgrounds In case of the proposed application of DHGR as NUI we have achieved an efficiency of 95.1%. The results obtained at various stages are promising and indicate the high applicability in real-time implementation in a DHGR system.

Biography

Richa Golash has completed her PHD. in 2022 from RGPV University Bhopal, India. She is working as research associate in Bennett University. She has teaching experience of 15 years and have deep interest in the field of artificial intelligence. She has published many research papers in various International/ National Journals/ Conferences.