尺度空间特征下人机交互多姿态三维手势智能识别OA
Intelligent multi-pose three-dimensional gesture recognition for human-computer interaction under scale spatial feature
为应对人机交互中手势姿态多样性、尺度变化及复杂背景干扰等问题,提出一种基于尺度空间特征的人机交互多姿态三维手势智能识别方法.首先构建三维手势点云并转换为二值体素网格,结合金字塔多尺度结构与SIFT描述子提取具有空间分布特性的手势特征;其次利用三维卷积网络回归关节点热图实现精确定位,引入时间移位模块与LSTM网络对手势动态序列进行建模,实现多姿态手势实时智能识别.实验结果表明,所提方法对10类交互手势的综合识别置信度最高达99.68%,在虚拟游戏、办公与教学三类场景中的识别稳定性为97.7%、96.38%、98.67%,抗干扰能力为94.99%、93.85%、95.98%,可实现高精度、多姿态三维手势智能识别,为人机交互与虚拟现实应用提供可靠、自然的手势交互支持.
In allusion to the diversity of gesture postures,scale variations,and the interference of complex backgrounds in human-computer interaction,a method of intelligent multi-pose three-dimensional gesture recognition for human-computer interaction under scale spatial features is proposed.The three-dimensional gesture point cloud is constructed and converted into a binary voxel grid.The gesture features with spatial distribution features are extracted by combining the pyramid multi-scale structure and the SIFT descriptor.The three-dimensional convolutional network is used to regression the heat map of the joint points to realize the precise positioning,and the temporal shift module and the LSTM network are introduced to model the dynamic sequence of gestures,so as to realize the real-time intelligent recognition of multi-pose gestures.The experimental results prove that the comprehensive recognition confidence level of this method for 10 types of interactive gestures can reach up to 99.68%.In the three scenarios of virtual games,office work and teaching,the recognition stability is 97.7%,96.38%and 98.67%,and the anti-interference ability is 94.99%,93.85%and 95.98%,which can achieve high-precision multi-pose three-dimensional gesture intelligent recognition,and provide reliable and natural gesture interaction support for human-computer interaction and virtual reality applications.
肖锟;郭伶凤;敖思魁;吴维
新疆师范大学,新疆 乌鲁木齐 830000仲恺农业工程学院,广东 广州 510220仲恺农业工程学院,广东 广州 510220仲恺农业工程学院,广东 广州 510220
信息技术与安全科学
三维手势识别尺度空间特征关节点热图多姿态手势人机交互LSTM网络体素网格时序建模
3D gesture recognitionscale spatial featurejoint point heat mapmulti-posture gesturehuman-computer interactionLSTM networkvoxel gridtemporal series modeling
《现代电子技术》 2026 (6)
189-193,5
2022年度广州市社科规划课题(2022GZGJ313)2023年度普通高校重点科研平台和项目(2023ZDZX402)
评论