Research On 3D Endoscope Point Cloud Stitching Algorithm Based On Binocular Camera

Since the 21st century,with the rapid development of medical imaging technology,the imaging quality of endoscopic images is getting higher and higher,and doctors can obtain more and more information.Three-dimensional endoscopes can provide doctors with in-depth information,allowing doctors to efficiently complete operations,reducing patient pain and improving the success rate of surgery.This paper mainly completed the construction of 3D endoscope system,the calibration of binocular camera,the visualization of point cloud data,and obtained the rigid body transformation matrix based on the two-dimensional image feature point matching algorithm.Finally,for the stitching of multiple images,the point cloud stitching method based on key frame filter was proposed.The specific work is as follows:1.In this paper,binocular CMOS,computer,stomach model,guide rail,rotary table,two circuit boards and external light source were used to build a set of three-dimensional endoscope system to complete the follow-up experiments of point cloud stitching.2.The 3D endoscope system was calibrated by using the camera calibration function in the OpenCV library to obtain the internal and external parameters of the 3D endoscope.The 3D coordinates were obtained by using the 3D measurement program based on SGBM method in the laboratory.3.In this paper,a method of obtaining rigid body transformation matrix based on 2D image feature points is proposed to achieve rough matching of point clouds.Firstly,this method use SURF algorithm to extract the image feature points,use FLANN methods and preliminary feature point selection based on Euclidean distance,use RANSAC method again to remove the image matching by mistake,and combined with the removal characteristics of 3D point cloud information without a corresponding point for 3D information.eventually,the method extracts the feature points of three groups of highest similarity by using SVD method,The rigid body transformation matrix is estimated.The transformed point cloud is used for ICP stitching.Compared with ICP method without coarse registration of point cloud,the efficiency is 5.1 times that of the latter.4.For 3D stitching of multiple images,a point cloud stitching method based on key frame filter is proposed.This method fully considers various camera movements that may occur in the actual scene of the endoscope,such as translation,rotation and scaling.It has strong robustness,and this method can effectively reduce the running times of ICP algorithm.The experimental results show that the stitching of 15 image sequences by the method in this paper can increase the field of view of three-dimensional stitching from 37.25° to 76.28°.the field of view has an increase of 2.1 times.the root mean square error of point cloud stitching region is 1.03.The operating speed is 4.1 times higher than that of the traditional ICP method.