• •

### 基于IASPP-ResNet分割算法的手势识别

1. 1.河北大学网络空间安全与计算机学院，河北 保定 071002
2.河北省机器视觉工程研究中心，河北 保定 071002
• 收稿日期:2020-12-31 修回日期:2021-03-27 出版日期:2022-07-04
• 通讯作者: 崔振超
• 作者简介:雷玉 女，1995年生，山西晋中人.河北大学网络空间安全与计算机学院硕士研究生，研究方向为深度学习、图像分割技术及其应用.
崔振超（通讯作者） 男，1983年生，河北邯郸人，讲师.中国计算机学会会员，2007年于燕山大学获得学士学位，2010年于燕山大学获得硕士学位，2015年于哈尔滨工业大学获得博士学位.现为河北大学网络空间安全与计算机学院教师，主要从事人工智能、机器视觉方向研究. E-mail: cuizhenchao@gmail.com
陈丽萍 女，1974年生，河北保定人，讲师.1997年于河北农业大学获得学士学位，2000年获得硕士学位.现为河北大学网络空间安全与计算机学院教师，主要从事机器视觉方面的研究.
陈向阳 女，1977年生，河南三门峡人，讲师.2000年毕业于燕山大学获得学士学位，2007年毕业于河北大学获得硕士学位.现为河北大学网络空间安全与计算机学院教师，研究方向为深度学习.
王煜骁 男，1997年生，河北廊坊人.河北大学网络空间安全与计算机学院硕士研究生，研究方向为深度学习、图像分类.
• 基金资助:
河北省自然科学基金项目(F2017201069);河北省研究生创新资助项目(HBU2021ss059)

### Hand Gesture Recognition Based on IASPP-ResNet Segmentation Algorithm

LEI Yu1,2, CUI Zhen-chao1,2(), CHEN Li-ping1, CHEN Xiang-yang1, WANG Yu-xiao1

1. 1.School of Cyber Security and Computer，Hebei University，Baoding，Hebei 071002，China
2.Hebei Machine Vision Engineering Research Center，Baoding，Hebei 071002，China
• Received:2020-12-31 Revised:2021-03-27 Online:2022-07-04
• Contact: CUI Zhen-chao

Abstract:

Gesture recognition is an essential research area in the field of computer vision, and it is also a significant component of the human-computer interaction. Due to its recognition results can be influenced by complex backgrounds, gesture recognition faces huge challenges. To solve the problem that is affected by the complex background, this paper proposes a new gesture recognition algorithm based on the combined model of dense segmentation and gesture classification. In the dense segmentation part, this paper shows the Improved Atrous Spatial Pyramid Pooling (IASPP). IASPP is a pooling layer in a convolution neural network, which can obtain the refine features by connecting cascade model and parallel model in atrous spatial pyramid pooling. Otherwise, in order to improve the segmentation performance by integrating details and spatial location information at different levels, the IASPP was embedded in a ResNet with encoder-decoder, and we name the method the Improved Atrous Spatial Pyramid Pooling-ResNet (IASPP-ResNet) for gesture segmentation. In the part of gesture recognition, we use the deep convolutional neural network model to obtain a higher recognition rate. The experimental results show that the IASPP-ResNet segmentation algorithm has a higher accuracy rate on the commonly used public data sets, compared with the traditional machine learning methods as well as the deep learning-based methods, and the gesture recognition rate of the combined model of dense segmentation and gesture classification proposed in this paper can reach 98.63% on NUS-II dataset, which is superior to the existing gesture recognition algorithm.