Robust tracking of moving hand in coloured video acquired through simple camera

Abstract

Interaction through dynamic hand gestures is an interesting yet challenging area because in computer vision the hand is a non-rigid object that moves unpredictably. Additionally, the real-time backgrounds are unstable. RGB data of moving hand are sensitive to light variation and camera-view thus, continuous localisation of hand region in RGB images is strenuous. This paper proposes a unique solution that combines Scale-Invariant Feature Transform (SIFT) features with automatic feature extraction mechanism of Region-based Convolutional Neural Network (R-CNN), for robust tracking of moving hand in coloured video acquired through a camera with normal resolution. The efficiency of the proposed methodology is 96.84% in simple and 94.73% in complex background. The comparative analysis with contemporary techniques working on RGB images exhibits that the proposed solution gives high accuracy in a real-time environment. In the future, we can design an economical and user-friendly natural user interface using the proposed technique.