Design And Study Of The Vision Based Fall Detection System For The Elderly

With the acceleration of aging in today’s society,the proportion of the elderly is increasing.The health and safety problems of the elderly gradually become the focus of attention of the family and society.The fall of the elderly is one of the main causes of disability and death.When the elderly fall,they can not find and seek help in time and often make the consequences more serious.In view of the injury to the elderly,a device that can accurately detect the fall of the elderly is essential.The intelligent monitoring system based on computer vision has become a hot research topic for the fall of the elderly due to its merits,low cost and multi task parallelism.For most of the fall studies at this stage are based on PC-side recorded video,this paper presents a combination of Raspberry Pi and Open CV real-time video surveillance system.Two algorithms of improved fall detection based on human scale feature and fall detection based on HOG and SVM are proposed and validated on this system.The main contents of the full text include:(1)In order to study the real-time fall,a monitoring system platform combining raspberry pie and OpenCV open source visual library was set up.The advantages and disadvantages of raspberry pie and OpenCV were introduced.The software configuration and other related operations were also carried out,and the commonly used image processing methods were studied.(2)An improved fall detection algorithm based on the proportion of human body is proposed.The algorithm uses the principle of moving target detection,using three human shape features to detect falls.Body aspect ratio effectively exclude the daily activities of the general,the effective area ratio such as stretching the arm to exclude the action,the center of the effective elimination of squatting and other movements.Finally,the algorithm is experimentally verified through the system platform.Experiments show that the use of a combination of three characteristics of the fall algorithm can effectively reduce the false positive rate and improve the fall detection accuracy.(3)A fall detection algorithm based on HOG features and SVM is proposed.The algorithm is based on machine learning.Firstly,the HOG feature extraction is introduced,and the sliding window method is used to detect the image.The algorithm uses a two-level SVM classifier.The first-level SVM classifier excludes most of the daily activities.The second-level SVM detects the fall.Finally,the algorithm is validated on the Raspberry Pi platform.The fall experiment shows that the fall algorithm based on the proportionality of the human body is sensitive to light,and the intensity of the light will affect the accuracy of the algorithm.The algorithm based on HOG features and SVM has high requirements on samples and systems.However,in summary,both algorithms have higher accuracy and lower false positive rate,which can meet the requirement of real-time detection of falls.