基于改进ViBe算法的双流CNN跌倒检测  

Two-stream CNN fall detection based on improved ViBe algorithm

作　　者：戚亚明 陈树越[1] 孙磊[1] Ya-ming;CHEN Shu-yue;SUN Lei(School of Microelectronics and Control Engineering,Changzhou University,Changzhou 213164,China)

机构地区：[1]常州大学微电子与控制工程学院,江苏常州213164

出　　处：《计算机工程与设计》2023年第6期1812-1819,共8页Computer Engineering and Design

基　　金：江苏省科技厅社会发展基金项目(BE2018638)。

摘　　要：为解决冗余信息导致跌倒检测准确率低的问题,构建一种基于改进ViBe算法的双流CNN网络模型来检测跌倒。将帧差法与ViBe算法相结合,解决原始ViBe算法的鬼影问题,通过多帧合成运动历史图(MHI)的方式输入双流网络,提出Sum-SoftMax决策层融合算法,增强算法模型的跌倒检测能力。在公开数据集上实验并与现有算法对比,实验结果表明,该模型综合准确率达到了98.7%,高于现有算法模型,可用于室内人员的跌倒检测。To solve the problem of low fall detection accuracy caused by redundant information,a two-stream CNN network model based on the improved ViBe algorithm was constructed to detect falls.The frame difference method was combined with the ViBe algorithm,the ghosting problem of the original ViBe algorithm was solved.Motion history image(MHI)was inputted to the two-stream network,and the Sum-SoftMax decision layer fusion algorithm was proposed,the fall detection capability of the algorithm model was enhanced.Through experiments on public datasets and comparison with existing algorithms,experimental results show that the comprehensive accuracy rate of the model reaches 98.7%,which is higher than that of the existing algorithm models.It can be used for indoor personnel fall detection.

关 键 词：跌倒检测 前景提取 鬼影去除 运动历史图 双流网络 卷积神经网络 网络融合 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]