Research On Fall Detection Of The Elderly Based On Dual Cameras

With the continuous replacement of the process of social aging,China will enter a moderately aging society before 2025.In an aging society,with the increase of the age of the elderly,their bones,muscles and other physiological functions begin to deteriorate,their senses become dull,their eyesight becomes blurred,their physical coordination becomes poor,and various accidents occur frequently.How to ensure the health and safety of the elderly has become an important issue in the society.Most of the old-age care models in China are home-based care for the elderly alone.For the elderly living alone,falling accidents have become a very serious problem.After the elderly fall,it is very easy to cause serious physical injuries,such as stroke and fracture.Falling to the ground for more than 1h will cause higher mortality.Therefore,the research on rapid response assistance for fall detection in the elderly is very necessary and of great significance.In order to solve the shortcomings of the existing fall detection model,such as low accuracy and lack of special fall detection data set.Combined with the current situation of aging and aging in China,that is,the research status of fall detection,taking the robot and real human body as the main body,this paper establishes fall detection models based on different data sets and different algorithms.This research uses the interdisciplinary knowledge of optics,auto-control and other disciplines to innovate the algorithm and hardware system,mainly including:(1)Establish the attitude data acquisition platform of humanoid robot,and create a mobile fall detection model.The robot posture data set is collected and analyzed by dual host computers,and its posture classification is mapped to the posture classification of the elderly.A fall detection model based on support vector machine is established by proposing an algorithm integrating the dual features of direction histogram and gray level co-occurrence matrix.The research not only solves the safety problem of falling data acquisition for the elderly,but also solves the problem of dead zone in traditional video surveillance.(2)This paper studies and establishes the human posture data acquisition platform,puts forward the method of processing the human posture image by using the key point detection technology,and then establishes the transfer learning fall detection model based on the above processing.Compared with the traditional fall detection model,the model established in this study has high accuracy,strong protection ability and better robustness.(3)A human fall detection model based on dual cameras is established.The traditional monocular camera model is used for detection during the day,and the infrared thermal camera human fall detection model is used for detection at night.By adding human tracking control system and continuous fall system,on the basis of greatly improving the accuracy of human fall detection,the false detection rate of falls is effectively reduced.The hardware experiment shows that the human fall detection model based on dual cameras has high detection accuracy and good real-time performance.