基于显著性特征的多视角人体动作图像识别研究  

Research on multi⁃perspective human action image recognition based on significant features

作　　者：廖民玲[1] LIAO Minling(Guilin University of Electronic Technology,Guilin 541004,China)

机构地区：[1]桂林电子科技大学,广西桂林541004

出　　处：《现代电子技术》2024年第24期143-147,共5页Modern Electronics Technique

摘　　要：为了应对多角度、多姿态特点的人体动作图像识别难的问题,研究一种基于显著性特征的多视角人体动作图像识别方法。通过显著性区域检测模块处理多视角人体动作图像,获取显著性区域序列并拼接成多视角显著性区域拼接图;利用VGG-Net网络从中提取其卷积特征图,输入到CA(上下注意力)模块中,将有利于人体动作识别的区域予以突出显示,并输出人体动作类别标签的概率值,实现多视角人体动作识别。实验结果表明,所提方法能够有效识别多视角人体动作,同时通过整合显著性区域检测和CA模块,可以显著提升其在多视角人体动作识别方面的应用效果。In order to address the difficulty of recognizing human motion images with multiple angles and postures,a method of multi-perspective human motion image recognition based on salient features is researched.By using salient region detection module to process multi-perspective human motion images,a sequence of salient regions is obtained and concatenated into a multi-perspective salient region mosaic images.The convolutional feature map is extracted by means of VGG-Net network and input into CA(up and down attention)module.The region conducive to human action recognition is highlighted,and the probability value of human action category label is output to realize multi-perspective human action recognition.The experimental results show that the proposed method can effectively recognize multi-perspective human movements.By integrating salient region detection and CA modules,the application effectiveness of the method in multi-perspective human action recognition can be improved significantly.

关 键 词：显著性特征 多视角 人体动作 图像识别 类别标签 CA模块 LSTM单元 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]