Research And Implementation Of Intelligent Wheelchair Based On Deep Learning Estimation Method Of Head Pose

With the deepening of population aging and the number of disabled people increasing year by year,more and more people are facing the difficult situation of traveling,wheelchairs can meet the travel and other needs of people with inconvenient actions,however,the most widely used hand-pushing wheelchair and electric wheelchair cannot meet the travel needs of disabled people such as limb disability.Most of the existing smart wheelchairs are in the laboratory stage and the human-computer interaction method has problems such as being vulnerable to environmental interference,not in line with human behavior habits.Therefore,in this thesis,head pose estimation is studied,based on this,a smart wheelchair based on the head pose is designed and realized.The main research work is as follows:(1)At present,most of the research on head pose estimation needs to first carry out face detection,and then perform head pose estimation,aiming at the high coupling degree between the two stages,easy to cause cumulative error and low robustness,this thesis proposes a fusion algorithm of face detection and head pose estimation.Based on the YOLOv3 algorithm,the algorithm in this thesis improves the Feature Pyramid Networks so that it can perform face detection and head pose estimation at the same time,and the corresponding loss function is designed.At the same time,the Depthwise Separable Convolution is used for lightweight processing to reduce the amount of computation,so that it can meet the real-time requirements of embedded devices after model inference and Tensor RT acceleration.Through experimental comparison,it can be seen that the errors of the algorithm designed in this thesis are 6.52° and 6.94°respectively in pitch and yaw,and it can simultaneously detect the face and estimate the head pose,which proves the effectiveness and accuracy of the algorithm.(2)Aiming at the stability,safety and diversity of control methods of intelligent wheelchairs,this thesis designs a variety of intelligent wheelchair control methods,including head pose control,self-balance control and joystick control.The head pose control is based on the head pose estimation results,and the influence of head attitude estimation error is reduced by mean filtering,and the pitch angle and yaw angle are the control signals of front and rear direction control and turning,respectively.At the same time,the pressure sensor is used to recognize the intention of head pose control to avoid restricting the user’s head movement.The self-balancing control is based on the original information of the MPU6050 sensor,and after the attitude settlement through the quaternion algorithm,the serial velocity loop and vertical loop are used to keep it stable at all times.Finally,this thesis retains the traditional rocker control,and designs the rocker control system based on the rocker potentiometer.The test results show that the control system designed in this thesis has strong robustness and effectiveness.(3)Based on the above two research results,the hardware and software of intelligent wheelchair are designed,and the whole intelligent wheelchair system is realized.The intelligent wheelchair system mainly includes remote control board and motherboard,with STM32 microcontroller as the core,with motor drive circuit,sensor circuit and auxiliary circuit.After testing,this intelligent wheelchair has high robustness and practicality.