Intelligent System For Detecting Abnormal Behaviour Of The Elderly Based On Machine Vision

With the gradual aging of China’s population structure,the issue of elderly care is gaining more and more attention,especially the safety monitoring of the elderly is crucial.At present,most of the elderly institutions use the traditional manual monitoring means,which consumes a lot of human and financial resources and is also prone to problems such as untimely monitoring.With the emergence of intelligent elderly care,automatic abnormal behavior detection system based on machine vision and artificial intelligence is gradually developing.In this thesis,based on the investigation of domestic and foreign abnormal behavior detection methods,an improved object detection and tracking algorithm is constructed to design and implement three abnormal behavior detection methods for elderly people in nursing homes and build the system,which is of certain practical significance for building a safe and intelligent nursing care environment.The thesis encompasses the following sections that outline the specific research work conducted:1.Based on the YOLOv5 object detection algorithm,several improvements are proposed in this thesis.Considering the problem of small target detail feature loss caused by multiple downsampling of the network,the P2 small target detection layer is introduced to solve the problem by changing from three-scale detection to four-scale detection,strengthening the fusion of features between the high and low layers,and effectively improving the accuracy of small target detection without affecting the detection of large targets;To enhance the precision of object detection,the network incorporates the attention mechanism module,aiming to further improve the accuracy,three attention mechanisms are added separately and the network performance is compared to determine CBAM as the optimal attention mechanism module;meanwhile,we simulate the scenario to create the homemade dataset,then use the homemade dataset to train and experimentally compare and analyze each model,and conclude that the YOLOv5-P2-CBAM model,which incorporates the small object detection layer and the attention mechanism,is the best model.The accuracy and recall rates are well improved.2.In order to obtain more accurate and stable multi-object tracking results,the YOLOv5-DeepSORT object tracking algorithm is designed based on DeepSORT,and the object detector of the original algorithm is replaced with the improved YOLOv5P2-CBAM model to construct the YOLOv5-DeepSORT object tracking algorithm.And for this research scenario,the homemade tracking dataset is used to train the deep appearance model,and finally the accuracy and precision of the tracking algorithm in this thesis are improved through experimental comparison and analysis.On this basis,the corresponding recognition algorithms are designed for three abnormal behaviors:raising hands to actively ask for help,falling down and staying still for a long time,and the automatic detection and alarm of the three abnormal behaviors are realized.3.Combined with the actual needs,the YOLOv5-DeepSORT algorithm is used to detect and track the target,on the basis of which an intelligent detection and alarm system for abnormal behaviour of elderly people in nursing homes is built,realising the functions of real-time monitoring,abnormal behaviour detection and alarm,and alarm record query.Through the actual abnormal behavior video test,the real-time and accuracy of the system is evaluated,and the detection accuracy of the system is high and the delay time is low,which meets the requirements of daily use and can provide effective technical support for the safety monitoring of the elderly,and has good application prospects.