Multiview Hand Gesture Recognition using Deep Learning
Mallika Garg, Pyari Mohan Pradhan, Debashis Ghosh · 2021 IEEE 18th India Council International Conference (INDICON) · 2021
Gesture recognition is a challenging research topic since different gestures have different sizes, poses, and sometimes face occlusion. In this paper, we tackle this problem by utilizing multiview gestures for recognition. We propose a hand gesture recognition system that learns from multiview gestures in a Convolutional neural network (Convnet) to tackle the problem of self-occlusion. Extensive experiments are performed on isolated gestures for different possible combinations of multiview training and testing sets. The proposed multiview recognition system is evaluated over the HGM-4 dataset. Evaluations are performed based on the system accuracy and found that as the number of views increases, the system more accurately recognizes the gestures. Also, our technique outperforms other methods with an overall accuracy of 94.71% on HGM-4 dataset.