Research On Intelligent Monitoring System And Behavior Recognition For The Elderly

The rapidly developing intelligent monitoring system has met people's needs in the aspect of guaranteeing social security,family safety and production safety.It can realize not only real-time monitoring of dangerous scenes,but also automatic early warning of accidents.In the trend of population aging,the intelligent monitoring system for the aged has been deeply studied,but most of the system applications and research aim at a specific monitoring area and monitoring mode.Hence,the system still has some limitations in terms of intellectualization,convenience and comprehensiveness.In the context of the aged have great demand for guardianship and the existing system has defects,this paper studies the following three items.Firstly,designing and implementing a fully functional monitoring system.The proposed intelligent monitoring system is composed of four parts,including monitoring terminals,monitoring nodes,monitoring platform and user interaction.The monitoring terminal refers to information acquisition devices such as visual sensors,gas sensors,temperature and humidity sensors,which are used to provide residential environment information,fitness data,visual images,real-time videos and so on;monitoring nodes include Raspberry Pi monitoring nodes and wearable devices,which are used to provide data collection,basic information analysis,communication transmission and other functions;monitoring platform consists of cloud server,cloud database and cloud object storage,which are used to provide intelligent analysis,image storage,information storage and other functions;user interaction mainly provides web page interaction and short message early warning.Secondly,proposing an intelligent monitoring system using a distributed architecture to facilitate expansion in different application scenarios.There are multiple monitoring nodes with data analysis function in a monitoring system.The system mainly uses the information collected by wearable devices,gas sensors and visual sensors to analyze whether the aged under guardianship have potential life danger.The monitoring terminal will generate a mass of data,and the analysis function born by the monitoring nodes in the distributed system can effectively alleviate the pressure of the monitoring platform.Meanwhile,multiple monitoring nodes in the monitoring system have the advantage of information complementation.For example the multi-view visual images acquired by nearby nodes can enhance the accuracy degree of behavior recognition,thus ensuring the accuracy of the system in behavior recognition.Thirdly,in order to further explore the role of visual information,this paper makes improvement in the behavior recognition algorithm.At present,the behavior recognition algorithm based on convolution network has good effect,but the input of this method is static image,it is impossible to obtain more video temporal information.The dynamic image integrates multiple video frames and contains more temporal information.Therefore,the proposed improved algorithm combines the concept of dynamic image and uses it as the input of temporal segment networks(TSN).Meanwhile,in order to allow full play to the information complementary advantage of monitoring nodes in the distributed system and enhance the accuracy degree of behavior recognition,this paper proposes a fusion decision method based on Borda from multi-view,which makes a synthetic judgment of the recognition results from multiple perspectives.