Research On The Tracking Method Of Human Sitting Position Of Toilet Assistant Robot Based On Depth Vision

In the current disability of the elderly,the phenomenon of the disabled toilet behaviors of the elderly is more obvious.In the current medical care,it is necessary to study the dynamic interaction between the disabled elderly and the toilet assistant robot to reduce the behavioral burden of the disabled elderly and the nursing staff.Therefore,it is of great significance to study the vision detection and seat position prediction algorithm of the toilet assist robot to solve the toilet problem of the disabled elderly.This paper first introduces the characteristics of toilet assistant robot and D435 depth vision sensor.Through D435 depth vision sensor,the depth image of the mixed standing posture distribution type of the nursing staff and the user can be obtained.The target detection classification framework is used to classify the different two person mixed standing posture.Then,the pixel positions of the user’s hip and knee joint points distributed in the two person mixed standing posture are obtained through the human joint point estimation framework,and the world distance information between the user’s hip joint and knee joint and the robot is obtained.The prediction algorithm is used to predict the sitting position of the user’s hip joint.Drive the toilet assist robot to move to the user’s predicted sitting position,thereby reducing the action burden of nursing staff and users.This paper mainly obtains the distance information of the user’s hip joint and knee joint through the depth camera of D435.The Kinect depth sensor joint point estimation algorithm will offset the joint point estimation of the two person hybrid standing posture,resulting in the inability to correctly estimate the pixel coordinates of the user’s hip joint and knee joint.To solve this problem,the YOLOv4-Tiny target detection framework is proposed to classify the standing posture distribution of nurses and users.The pixel coordinate positions of the user’s hip joint and knee joint cannot be estimated after the target detection classification is completed.In this paper,an attitude estimation algorithm based on Deep Lab Cut is proposed to train the corresponding standing position images.The pixel coordinate information of the user’s hip joint and knee joint is estimated through the obtained training posture model.Finally,the effectiveness of joint point estimation is verified by the training results of Deep Lab Cut.Obtain the pixel coordinates of the hip joint and knee joint of the two person mixed standing posture,and use the pixel coordinates of the d435 software development kit to convert the world coordinate system to obtain the distance between the human joint point and the robot.Facing the problem of predicting the user’s sitting position,this paper uses artificial neural network to train the motion model of each sitting position and establish the extended Kalman filter motion equation.The extended Kalman filter motion equation of each sitting trajectory is combined with the multi-model interaction algorithm to predict the user’s sitting position.After obtaining the information of the sitting position,the toilet assistant robot aims at the situation that the posture of the robot is deflected due to external interference such as friction during the movement of the robot.The dynamic model of the toilet assistant robot is established,and the sliding mode controller is designed.Ensure that the posture deflection of the toilet assist robot is within a reasonable range in the process of tracking the user’s sitting position,and complete the toilet needs of the disabled elderly.Through the human-computer interaction experiment of the toilet assistant robot,the effectiveness of the scheme of tracking the sitting position proposed in this paper is proved.