| With the continuous growth of the elderly population in China,the health problems of the elderly have also been widely concerned by the society,especially the elderly living alone have the problems of taking the wrong medicine,taking medicine indiscriminately and some abnormal behaviors that cannot be found in time.In this paper,an intelligent medicine cabinet system is designed to solve the problem that the elderly take the wrong medicine and take medicine indiscriminately,and the abnormal behavior is identified through the information such as environment,signs and videos collected by various sensors and cameras on the intelligent medicine cabinet,and the fall detection technology based on computer vision is emphatically studied.The main research contents of this paper are as follows:(1)An intelligent medicine cabinet is designed and implemented to solve the problems of the elderly taking the wrong medicine and taking medicine indiscriminately.The system is divided into intelligent medicine administration subsystem,data acquisition subsystem,abnormal behavior detection subsystem and data display subsystem.The intelligent drug delivery system can regularly remind users to take medicine,pump water and other operations,and the required drugs can be given only after the face recognition of the camera is successful.The data acquisition subsystem adopts STM32 single-chip microcomputer as the main control chip,and designs and manufactures an information acquisition device through a 3D printer,which can collect and analyze information such as temperature,humidity,heart rate and blood pressure in real time and find abnormalities in time.Display the collected data on the mobile phone through the We Chat applet to facilitate the guardian to check the situation of the elderly,so as to learn more about the situation of the elderly.(2)Aiming at the problems of environmental interference and visual changes,the target detection algorithm based on YOLOv5 is studied,and an improved detection algorithm for abnormal behavior of elderly people falling in YOLOv5 is proposed.ACON(Activate or Not)activation function is used instead of Si LU as activation function to improve the classification and detection effect of the model;By adding the coordinate attention mechanism,the object of interest can be located and identified more accurately;Ghost Conv is used as convolution module,and more feature maps are obtained by linear operation,which reduces the amount of parameters and calculation,and improves the accuracy of model detection.Experiments are carried out on the fall data set(URFD).The results show that the parameters of the model trained by the improved YOLOv5 algorithm are reduced by 35.14%,the model size is reduced by 4.8M,and the accuracy can reach99.1%.(3)Aiming at the problem that using a single YOLOv5 to detect falls is easily influenced by similar behaviors,a method based on lightweight Open Pose attitude estimation and improved YOLOv5 is further adopted to detect falls.Firstly,the human joint information is extracted from the video image.Then,the descending speed of the human center point is calculated based on the extracted joint information,and the falling behavior is detected under a given threshold according to the aspect ratio of the human circumscribed rectangle.The experimental results show that the method adopted in this paper has achieved good results in human fall detection.In this paper,an intelligent medicine cabinet management system is designed and implemented,which provides services such as regularly reminding to take medicine,automatically giving medicine,finding abnormality and warning,and studies the video abnormality detection technology based on deep learning.An improved YOLOv5 algorithm is proposed,which reduces the model size and parameters and improves the accuracy of anomaly detection.Furthermore,a method combined with lightweight Open Pose attitude estimation is proposed to solve the problem of false detection of similar behaviors.The work in this paper can provide a perfect system for community service stations,nursing homes and hospitals,and the proposed method can also provide theoretical support for human abnormal behavior detection. |
PDF Full Text Request