摘要
在复杂的动态环境中,受人群、背景干扰等因素的影响,会导致图像中出现大量的视觉噪声,从而干扰人体动作的准确检测。针对动态环境影响检测效果的问题,提出基于机器视觉的动态环境人体运动状态检测方法。结合广义总变分方法和自适应去噪方法建立融合方法去噪视频图像,在传统混合高斯算法基础上,提出自适应学习率混合高斯算法,并将其与五帧差分算法和形态学处理相结合,用于动态环境视频图像背景差分。采用鲁棒主成分分析法,提取视频序列低秩运动信息,并构建双流卷积神经网络,分别输入人体运动特征图像和低秩运动信息,将输出结果加权融合,实现动态环境下的人体运动状态检测。实验结果表明,所提方法去噪效果好、检测精度高,且检测所用时间短。
In complex dynamic environments,crowd interference and background interference may cause a large amount of visual noise in images,thus interfering with the accurate detection of human action.To address the issue of detection effectiveness affected by dynamic environments,this article put forward a method of detecting human motion states in dynamic environments based on machine vision.Firstly,a fusion method combining generalized total variation and adaptive denoising methods was used to denoise video images.Based on the traditional mixture Gaussian algorithm,a mixture Gaussian algorithm based on adaptive learning rate was proposed and combined with the five-frame difference algorithm and morphological processing to perform background subtraction of video images in dynamic environments.Then,robust principal component analysis was adopted to extract low-rank motion information from video sequences and build a two-stream convolutional neural network.Respectively,the human motion feature image and low-rank motion information were input,and then the output results were weighted and fused.Finally,the detection of the human motion state in dynamic environments was achieved.Experimental results show that the proposed method has a good denoising effect,high detection accuracy,and short detection time.
作者
刘玉洁
万瑜
LIU Yu-jie;WAN Yu(Chengdu College of University of Electronic Science and Technology of China,Chengdu Sichuan 610054,China;School of Life Science and Engineering,Southwest Jiaotong University,Chengdu Sichuan 610000,China)
出处
《计算机仿真》
2024年第12期570-574,共5页
Computer Simulation
基金
教育部2019年第二批产学合作协同育人项目(201902005043)
四川省教育厅自然科学一般项目(18ZB0256)
四川省第二批地方普通本科高校应用型示范专业项目(255-256)。
关键词
机器视觉
动态环境
人体运动状态
混合高斯算法
双流卷积神经网络
Machine vision
Dynamic environment
Movement state of the human body
Mixed Gaussian algorithm
Two-stream convolutional neural network
