基于浅三维稠密网的多模态手势识别算法  

Multimodal Gesture Recognition Algorithm Based on Shallow 3D Dense Networks

作　　者：邓智方 袁家政 刘宏哲 原春锋[3] 张宏源 DENG Zhifang;YUAN Jiazheng;LIU Hongzhe;YUAN Chunfeng;ZHANG Hongyuan(Beijing Key Laboratory of Information Service Engineering,Beijing Union University,Beijing 100101,China;Office of Academic Research,Beijing Open University,Beijing 100081,China;National Laboratory of Pattern Recognition,Institute ofAutomation,ChineseAcademy of Sciences,Beijing 100190,China)

机构地区：[1]北京联合大学北京市信息服务工程重点实验室,北京100101 [2]北京开放大学科学研究处,北京100081 [3]中国科学院自动化研究所模式识别重点实验室,北京100190

出　　处：《计算机工程与应用》2019年第19期166-172,共7页Computer Engineering and Applications

基　　金：国家自然科学基金(No.61571045,No.61871039,No.61871028);北京成像技术高精尖创新中心高精尖项目(No.BAICIT-2016002);北京市属高水平教师队伍建设支持计划项目(No.IDHT20170511);国家科技支撑项目(No.2015BAH55F03);北京联合大学研究生资助项目

摘　　要：手势识别旨在理解人体的动态手势,是人机交互领域极其重要的交互方式之一。该方法通过将二维稠密网扩展为三维稠密网,并加入Inception结构,提出了一种基于浅三维稠密网的多模态手势识别方法,将其命名为Spatial Temporal 3D(ST3D) dense network。所提出的方法在手势识别公开数据集大规模离散手势数据集(IsoGD)上进行了评估,并取得了目前最好效果。实验证明,所提方法能够有效地学习到视频样本中手势的短期、中期以及长期时空特征。Gesture recognition aims at understanding dynamic gestures of the human body,and is one of the most important ways of human-computer interaction.A multimodal gesture recognition method based on a shallow 3D dense network is proposed by extending the two-dimensional dense network into a 3D dense network and adding the Inception structure, which is named Spatial Temporal 3D(ST3D)dense network.The proposed method is evaluated on the Charlearn LAP large- scale Isolated Gesture Dataset(IsoGD)and achieves the best results.Experimental results show that the proposed method can effectively learn short,mid and long term spatiotemporal features of gestures in video samples.

关 键 词：ST3D方法 Inception结构 多模态 手势识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]