Research On Human Falling Detection Technology Based On 3D Vision

At present,China has begun to enter the aging society.At the same time,falling has become the main reason for elderly injury,especially for empty nest whose proportion is gradually increasing.In this paper,Kinect device is employed to realize human fall detection based on 3d vision.The main work is as follows:Firstly,a kinect-based fall detection system is proposed.It can move with human on a mobile platform combining Kinect and Mecanum wheel.According to the system scheme,the hardware and software of fall detection are designed.With the human tracking function of Kinect,the most real falling data of human bone point is collected by monitoring object in real time.During data acquisition and fall detection,the mobile platform always keeps a stable distance from the human body to obtain the best data.Secondly,Kinect bone tracking technology is used to obtain the depth image of the human body and extract the main key points of the human body.The distance between Kinect and human body is measured by Kinect,and the change of distance is applied to determine the movement direction of human.Then the data flow is converted into electrical signals so that can control the rotation angle of Mecanum wheel.The distance between the mobile platform and the human body is kept at a constant distance of 1.5m.Thirdly,Kinect bone tracking technology can be used to convert the detected depth image into 25 key nodes of human body and extract 3d coordinate information of key bone points.Then,feature points(head,shoulder,spine)are extracted according to the fall data.A plane perpendicular to the ground and an orthogonal section of the plane are established by the least square method with the extracted feature points.In addition,the distance between the main bone point and the two sides and the change of the speed are calculated,and the threshold method is used to determine whether the fall occurred.Fourth,Kinect 2d data and 3d data are combined to identify human objects.The human body is separated from the background in the RGB image obtained by Kinect,and the distribution of human RGB pixels in the figure is analyzed from the separated human body image.Object in different depth is identified on the basis of depth data containing the distance between Kinect camera and human in depth image.The depth data is used to segment and extract the target body in the image.When multiple target human bodies are at the same depth,the Kinect bone tracking function is used to obtain the information of bone points of human bodies in 3d space,and theEuclidean distance of bone points is calculated by using the different distribution of bone points of different human bodies in 3d space.The size of Euclidean distance is used as the identification basis for judging different human bodies.Experimental results show that the proposed algorithm performs well.