Research On Elderly Behavior Detection Algorithm Based On Multimodal Fusion

With the aging trend of the world population becoming more and more obvious,the aging society is getting closer and closer,the pension problem is becoming more and more serious.In recent two years,due to the rapid development of artificial intelligence technology,the social demand for intelligent products for the elderly based on artificial intelligence,mobile Internet,Internet of Things and other technologies is also growing.Among them,the daily life behavior of the elderly group is one of the key points that cannot be ignored in the development of intelligent products for the elderly.Because only by fully mastering the daily life behavior of the elderly group,can the intelligent products developed be more specific and precise to serve the elderly.However,according to the current research,there is still a big gap between the research on the daily life behavior of the elderly and people’s expectations.Especially in the actual application process,due to a series of uncontrollable factors such as environment,the identification accuracy is reduced;At the same time,the behavior data of the elderly studied is generally single,and there are few studies on the behavior data of the elderly with multiple sources and heterogeneity,and there is also a lack of relevant systematic methods.In terms of application background,few scholars have researched and developed products related to the elderly’s daily health scientific behaviors based on the identification of their daily life behaviors from the perspective of their potential needs.Therefore,this paper takes the elderly group as the research object,in line with the purpose of providing convenient life and scientific health for the elderly,so as to solve the pain and difficulty problems in the present stage of the elderly behavior identification model.This paper takes this topic as the research topic,explores the daily life behavior of the elderly,and provides theoretical basis and technical support for the development of intelligent products for the elderly.The innovation of this paper is mainly reflected in the following two aspects:(1)The goal of this topic is to explore the behavior of the elderly under robust and reliable algorithm.I don’t want to introduce a specific data set to cause errors in the real scene,so I created a data set of this topic during my graduate study.The image information and sensor information of the daily behavior of the elderly are collected at the same time to complete the behavior data set of the elderly,which provides data support for the multi-modal information fusion and the research on the behaviors with unique characteristics of the elderly.(2)An elderly behavior recognition algorithm based on multimodal fusion was proposed to construct LRCMN model.The network uses both the rich appearance feature information provided by RGB data and the acceleration information provided by the inertial sensor,and uses the feedforward neural network to complete the information fusion of the two modes on the mid-decision layer of the model,which enables the model to obtain the deep information of the multi-source data of the elderly,and share and complement the two modes of heterogeneous information in the fusion process.The algorithm has been verified on the dataset created in this project,and a good recognition rate has been obtained.At the same time,the model adopted the model parameters trained by two modes in the initial parameter stage,which is helpful to enhance the generalization ability of the model.