Research On Head Pose Estimation And Motion Tracking Based On Facial Landmarks

Transcranial magnetic stimulation(TMS)is a non-invasive treatment for brain diseases for the elderly.It possesses the advantages of simple operation and good therapeutic effect.Therefore,TMS treatment has been widely promoted and used in clinical treatment in many countries in recent decades.However,in the actual TMS treatment,it is difficult for the doctor to determine the location of the lesion,and the patient's head leaves the stimulation area of electromagnetic coil,which greatly reduces the treatment effect.Therefore,this thesis researches on the head pose estimation algorithm and motion tracking algorithm.In the thesis,the demand of TMS therapy is analyzed.The overall design scheme and performance index of TMS treatment robot system are given.Firstly,the position of the lesion point of the brain is obtained by stereotactic technology.Then the visual sensor and the force sensor are used to respectively detect the pose of patient's head and the contact force between the head and electromagnetic coil.Finally,the robot arm fixed with the coil is controlled to track the lesion point.In this thesis,an algorithm of head location is designed based on facial landmarks,which is combined with a depth camera to detect the position and pose of head.The Harrlike feature is used to construct a cascaded classifier in order to detect the face region from the video stream,and the local constraint model is used to detect the facial landmarks in the face region.Joint bilateral filter is used for depth image restoration.The depth and color images after alignment are used to obtain the head position and pose respectively,and the 3D spatial position of the lesion points is obtained.Besides,in this thesis,head motion tracking algorithm is studied.For the information obtained from the force sensor,the gravity of the tool on the end-effector is compensated,and a position controller of the end-effector is designed.The pose of the end-effector is adjusted according to the contact force,which can guarantee the contact force to be within a certain range.At the same time,a tracking controller is designed according to the spatial information of patient's head.The adaptive parameters are obtained in the controller to determine the speed of the end-effector.In the thesis,the hardware platform and algorithm platform of TMS treatment robot system are set up.The accuracy and the real-time performance of the system are verified through experiments.Besides,PC software is designed to realize the visualization of the lesion point localization and the tracking of the treatment process.