Multi-Target Tracking And Locating For Minimally Invasive Spine Surgery Training System

Spine disease is a common clinical disease,including disc herniation,bone hyperplasia,spinal stenosis and so on,which can not only cause physical pain for the patients,but also result in additional mental stress.Spine surgery consists of open spine surgery and minimally invasive spine surgery.Minimally invasive spine surgery is becoming a trend of surgical development and gradually evolving into modern medicine direction of development,due to the advantages such as small open cuts,less bleeding,less pain,less risk of preoperative and postoperative complications and so on.However,the facts such as technical difficulties,long training cycles,lack of training mechanisms have restricted the development of minimally invasive spine surgery.The pedicle screw implantation requires precise operations,whose accuracy can directly impact the surgery outcome.Hence,there is an immediate need for more research on minimally invasive spine surgery training system.In order to provide a more realistic training environment to the trainers,the training system should include precise visual and tactile feedback.Therefore,based on augmented reality,minimally invasive spine surgery training system with real force feedback is designed in this thesis.A virtual 3D interface providing trainers with realistic visual feedback is used to observe the relative position of surgical instrument regarding to the spine in all directions and angles.3D printed spine model and surgical instrument with reference frame are employed to give trainers real force feedback.The whole thesis is organized as follows.Firstly,according to the existing basis of the task group,based on augmented reality with real feedback,minimally invasive spine surgery training system is designed.The software and hardware structures of the system are introduced and the workflow of the whole training system is described.Secondly,a calibration algorithm based on plane checkerboard between traditional and adaptive calibration algorithm is used to calibrate the binocular camera to get the inside and outside matrices and distortion coefficients,which provides the basis for further threedimensional reconstruction.The mapping error of the binocular camera based on the checkerboard plane calibration algorithm is 0.12 pixel,which satisfies the accuracy requirement of three-dimensional reconstruction.Thirdly,given that the acquired images do not have to be of high quality,the median denoising algorithm is adopted to preprocess the raw images.The based on gray-morphological image edge enhancement algorithm is employed to improve the edge features of the targets.Then Canny algorithm is used to detect edges of the targets.To address the problem that Hough transform algorithm is not suitable for the extraction of moving targets in affine change situation,an affine moment invariant algorithm based on contour is proposed,which provides the data foundation for further three-dimensional reconstruction.Fourthly,after the stereo matching algorithm for multi-target feature points,three-dimensional reconstruction algorithm based on binocular camera is used to get the coordinates of the targets in the world coordinate system.Dynamic tracking of the position of the center of interest region based on Kalman filter is employed to improve the real-time performance of multi-target dynamic tracking and locating.Tracking algorithm is designed considering the case of information loss.The precision error of multi-target dynamic tracking and locating based on binocular camera is controlled within 1mm,which meets the accuracy requirements of minimally invasive spine surgery training system with real force feedback based on augmented reality.