摘要
为了提高多帧视频序列中动态手势的识别效果,结合计算机视觉分析,提出了基于卷积神经网络视觉融合的动态手势识别方法。采用模糊数据多频谱方法进行多帧视频序列中动态手势视觉图像采集,对采集的图像用Harris角点检测和多传感识别方法进行多模状态分层特征点标定,用卷积神经网络视觉融合方法提取多帧视频序列中动态手势动作的边界轮廓特征点信息,分析多帧视频序列中动态手势的层次化分割特征,用图像分割和边缘信息增强方法,提高动态手势图像的分辨能力,结合角点优化检测技术,用视觉动态跟踪分析实现对手势动作特征点的自动化标定,根据动态手势的帧点分布规则实现多帧视频序列中动态手势动作图像的自适应特征检测和识别。仿真结果表明,采用该方法进行多帧视频序列中动态手势识别的准确性较高、实时性较好。
In order to improve the effect of dynamic gesture recognition in multi-frame video sequences,a dynamic gesture recognition based on visual fusion of convolutional neural network is proposed combined with computer vision analysis.Adopt fuzzy data multi-spectrum acquisition method to collect dynamic gesture visual images in multi-frame video sequence,adopt Harris corner detection and multi-sensor recognition method to calibrate multi-mode state layered feature points,adopt convolution neural network visual fusion method to extract boundary contour feature point information of dynamic gesture actions in multi-frame video sequence,analyze hierarchical segmentation features of dynamic gestures in multi-frame video sequence,then the image segmentation and edge information enhancement methods are adopted to improve the resolution of dynamic gesture images.Combined with corner optimization detection technology and visual dynamic tracking analysis,automatic calibration of gesture motion feature points is realized,and adaptive feature detection and recognition of dynamic gesture motion images in multi-frame video sequences are realized according to frame point distribution rules of dynamic gestures.Simulation results show that this method has high accuracy and good real-time performance for dynamic gesture recognition in multi-frame video sequences.
作者
于海鹏
李博
王旭辉
YU Haipeng;LI Bo;WANG Xuhui(College of Computer,Henan University of Engineering,Zhengzhou 451191,China)
出处
《河南工程学院学报（自然科学版）》
2021年第3期63-67,共5页
Journal of Henan University of Engineering：Natural Science Edition
基金
河南省高等学校重点科研项目(20A520010)。
关键词
卷积神经网络
视觉
融合
动态手势
识别
多帧视频序列
convolution neural network
vision
fusion
dynamic gesture
identification
multi-frame video sequence
