Dense Reconstruction Of 3D Scene Of Airway Based On Endoscope Image Of Airway

During airway endoscopy,physicians can generally only get the two-dimensional image of adenoid,and are thus incapable of accurately obtaining the 3D structure of the patient’s adenoids.Furthermore,it is inconvenient to frequently change the angle of the endoscope during the inspection time,since it may make the patients quite uncomfortable.In addition,general airway endoscopy relies on the doctor’s subjective judgment of the degree of adenoid obstruction of the nostril,and heavily relies on the experience and judgment of the operating doctor,which lacks objectivity.This paper studies the dense reconstruction of the airway three-dimensional scene based on airway endoscopic images,and proposes a three-dimensional reconstruction method based on endoscopic images to reconstruct the adenoid area under the current endoscope field of vision.The main contents of this paper are as follows :(1)Summarize and analyze the advantages and disadvantages of the existing three-dimensional reconstruction technology,and point out the necessity of the three-dimensional reconstruction technology based on endoscopic images.(2)The geometric model of camera imaging and the mathematical principles of stereo vision are studied,and the three-dimensional reconstruction algorithm based on stereo vision is realized.This algorithm includes data acquisition and preprocessing,feature point matching,Fundamental matrix calculation and triangulation.Specifically,after preprocessing the video image data,the pose(rotation matrix,translation vector)between each frame image is obtained through the SURF operator matching feature points,and the result is uniformly optimized using Bundle Adjustment method to obtain a sparse three dimensional point cloud.(3)Optimize on the basis of the algorithm of(2)to complete further dense reconstruction.The optimization is mainly divided into three aspects.Firstly,we establish a matching point pair based on the correlation method,and we calculate the zero normalized cross-correlation(ZNCC)between two image patches.In the second place,we optimize the Fundamental matrix which is the description of epipolar geometry between the two images to calculate more accurate pose(rotation matrix,translation vector)between each frame of the image,and we use the optimized Fundamental matrix to narrow the matching range through the epipolar constraint to speed up operation speed;Last but not least,we define the energy function with depth as a variable,and optimize the depth value by setting the error term and smoothing term to further improve the accuracy of reconstruction.The experimental data used in this article are three sets of airway endoscopic video images from the Department of Otolaryngology,Shenzhen Hospital of Southern University of Science and Technology,and the reconstruction target is the adenoid area under the endoscope.The experimental results are analyzed from the matching results and the reconstruction results,and the feasibility of the three-dimensional reconstruction algorithm based on stereo vision is verified in the endoscopic image.