Human action and gesture recognition provides important and worth information for interaction between human and device ambient that monitors living, healthcare facilities or entertainment activities in smart homes. Recent years, there were many machine learning model application studies to recognize human action and gesture. In this paper, we propose a dynamic hand gesture recognition system in video based on two stream-convolution network (ConvNet) architecture. Specifically, we research the state-of-the-art approaches using to recognize dynamic hand gesture in video and propose an improvement method to enhance performance of model which is suitable for uses such as indoor environment in this paper. Our contribution is improvement of two stream ConvNet to achieve better performance. The results show that the proposal model improves execution speed and memory resource usage comparing to existing models.
TWO-STREAM CONVOLUTIONAL NETWORK FOR DYNAMIC HAND GESTURE RECOGNITION USING CONVOLUTIONAL LONG SHORT-TERM MEMORY NETWORKS
