Research And Application Of Abnormal Behavior Recognition For The Elderly Based On Deep Learning

Since the beginning of the 21st century,the aging population in our country has been increasing annually,and the problem of population aging has become prominent.Due to the ongoing shortage of elderly care professionals,inadequate facilities in elderly care institutions,and the difficulty of institutional elderly care to meet the needs of rural elderly care,home-based elderly care will gradually become the main form of elderly care in the future.Elderly people at home may experience falls,chest tightness,headaches,or long periods away from home,which are abnormal behaviors.If not discovered in time,these could cause more severe harm to the elderly.This work is based on a non-contact computer vision approach to recognize and classify the abnormal behavior of the elderly living alone by constructing a human detection model and a human abnormal behavior recognition model.and experimentally testing the performance of the elderly abnormal behavior recognition model.At the same time,the performance of the old abnormal behavior recognition model was tested by experiments.The main work is as follows:Firstly,based on YOLOv5,a human object detection algorithm is constructed to detect and locate human objects in the video input.To solve the problem of complex home environment and difficult detection of human body occlusion,an improved Inception module was added to YOLOv5 to extract feature information of different scales.At the same time,CBAM attention was added to the CSP structure,and EIo U loss function was used to improve the accuracy and generalization of the network to achieve accurate bounding box regression.Train the model on the preprocessed VOC2012 dataset to construct a human object detection model.Compared to the original YOLOv5 model,the accuracy and average accuracy of human object recognition have been improved by 2.5%and 3.1%,respectively.Then,a behavior recognition algorithm is constructed based on I3 D,and the feature fusion module(FFM)is used to fuse the Bo TNet and CA attention modules.The fused module is embedded into the I3 D network to improve the accuracy of the network’s behavior classification.Training and testing on the UCF-101 dataset showed that the model converged faster and improved recognition accuracy by 2.2%.At the same time,the three types of movements of the elderly,including falls,headaches,and chest pain,were defined as abnormal behavior of the elderly,and a video dataset of abnormal behavior of the elderly was created.The recognition accuracy of the elderly abnormal behavior recognition model trained on this dataset for the three types of abnormal behavior movements reached 93.4%,86.3%,and 83.1%,respectively.Finally,based on the improvement of the human object detection model and behavior recognition model mentioned above,YOLOv5 and I3 D were used to jointly identify the abnormal behavior of elderly people living alone.A set of abnormal behavior alarm system for elderly people living alone was designed and developed.The system can monitor the behavior of elderly people living alone through video or real-time monitoring.If the elderly person experiences abnormal behavior,it will immediately alarm and display abnormal behavior information.The experimental results indicate that the system has high accuracy and practicality.