Study On Head Modeling And Localization Method In Transcranial Magnetic Stimulation Treatment System Based On Vision

Transcranial Magnetic Stimulation(TMS)has become the most effective method for preventing and treating brain diseases for old people.At present,transcranial magnetic stimulation therapy is still performed by a doctor or a nurse in the clinic.The transcranial magnetic therapy coil is placed on the lesion point of the patient's head.It's important that the patient's head would be kept still during treatment.This causes the patients to lose freedom of movement and they have to keep a posture for a long time and it will bring a very bad experience to the patient.TMS therapy robot system is receiving more and more attention,in which the location and tracking of head-lesion point is one of the relatively important problems.If we have a patient's head appearance model,the doctor can indicate the position of the lesion from the head model.And we can track the lesion locations by tracking the head during the treatment.It is also convenient for the registration between the MRI skull model and the head appearance model.Kinect2 is adopted as the visual sensor and UR5 mechanical arm is used as the actuator.The methods of acquiring and locating the head is studied.The main research contents of this thesis are given as these followings.We analyze the needs of the TMS therapy robot system first.Then the overall composition of the robot treatment is introduced.The visual sensor for acquiring and locating the head model is determined and selected.And the hand-eye calibration between the Kinect2 and UR5 robot arm is performed.To solve the hole problem of Kinect2's depth image,the joint bilateral filtering algorithm is studied.The depth image and RGB image of Kinect2 are used to obtain the 3D point cloud.Based on the depth image of Kinect2,we use the ICP algorithm to obtain the head model and it is visually display using the point cloud library.Because the ICP algorithm is time consuming,the GPU is used to open up multiple threads for parallel computing,which improves the speed of the model acquisition.To solve the problem of patient's head movement during the treatment,a method of positioning the head using facial feature points is proposed.Firstly,the HOG feature descriptor and the support vector machine classifier(SVM)are used to detect the face,and the cascade regression algorithm ERT is used to locate the face feature points in the detected face region.And then using the Kinect2's aligned depth image,we get the spatial three-dimensional coordinates of the facial feature points.A few frames may be absent sometimes,and the facial feature points may be not stable.We use the sliding window filtering algorithm to solve the above two problems.Finally,by using the UR5 robot arm as the actuator,we can implement the head tracking task by identifying the facial feature points in real time.