文章基本信息

标题：Hand Gesture Recognition Based on Convolutional Neural Network
本地全文：下载
作者：M.Anitha ; V.Bala Raviteja ; Y.Lakshmi Kumari 等
期刊名称：International Journal of Early Childhood Special Education
电子版ISSN：1308-5581
出版年度：2022
卷号：14
期号：2
页码：6090-6095
DOI：10.9756/INT-JECSE/V14I2.694
语种：English
出版社：International Journal of Early Childhood Special Education
摘要：Humans and machines both use hand gestures to communicate. Vision-based dynamic hand gesture detection has grown in popularity as a research topic because of the wide range of applications it may serve. This research presents a novel deep learning network for recognising hand gestures. In order to learn both short-term and long-term features from visual inputs, the network mixes numerous well-proven modules, avoiding heavy computing in the process. In order to learn short-term features, each visual input is separated into a fixed number of frame groups. An RGB image and an optical flow snapshot are generated at random for each frame in the group. These two elements are merged and input into a convolutional neural network in order to extract features (ConvNet). There are identical settings for each ConvNet in a given group. Finally, a final classification result is predicted by an LSTM network, which processes all ConvNet output and learns long-term characteristics. Using the Jester and Nvidia datasets, which are both popular hand gesture datasets, the novel model was tested. Compared to other models, our results were very similar. Expanded hand movements were employed to illustrate the new model's robustness in the larger datasets.
关键词：Convolutional neural network (ConvNet);hand gesture recognition;long short-term memory (LSTM) network;shortterm sampling;transfer learning