3D Texture Reconstruction Of Abdominal Cavity Based On Monocular Vision SLAM

Most of the laparoscopic images of the abdominal cavity are displayed on a 2D navigation screen,which lacks intuitive three-dimensional spatial information,while the monitor is far away from the surgical area easily causing the problem of oculo-hand disorder for the surgeon.In order to better assist the surgeon in making accurate surgical judgments,this thesis is of great significance to minimally invasive surgery by investigating the 3D texture building of the abdominal environment in minimally invasive surgery,which can solve the problem of positioning and navigation of medical devices and abdominal lesions.For the narrow,humid environment,lack of features and repetitive abdominal environment,this thesis constructs a monocular SLAM system model,establishes the mathematical relationship of point projection from 3D space to 2D image plane by analyzing the imaging model of monocular laparoscope,uses the initialization of monocular SLAM system to provide the initial value of camera pose and combines the pose solving principle to obtain the coordinates of abdominal space point and laparoscope pose.To address the problem that the abdominal image has more specular reflective areas,this thesis introduces the CLAHE algorithm for abdominal image pre-processing to enhance the contrast between the blood vessels and the background.Due to the higher repeatability of the AKAZE algorithm for extracting feature points,the AKAZE-ORB algorithm is designed in this thesis based on the ORB algorithm,which is combined with the AKAZE algorithm for improvement.The algorithm uses the AKAZE algorithm to track the feature point orientation of the abdominal cavity image,combined with the binary descriptors of BRIEF for feature description.The algorithm improves the number of feature points detected and the matching accuracy in an abdominal cavity environment with highly similar features.To address the problems of information redundancy and cumulative errors in the construction of 3D maps of the human abdominal cavity,this thesis adjusts and optimises the key frames for constructing the positional maps by BA to improve the accuracy of laparoscopic tracking and the denseness of the reconstruction.In this thesis,the traditional Bo W model is improved by extracting the visual features of medical images in a way that generates a Bo W model specifically for medical images,which effectively reduces the time for solving the similarity of abdominal images and improves the robustness of closed-loop detection in real time.The point cloud densities of the maps constructed by the classical SLAM system are still sparse and cannot establish a comprehensive intuitive description of the abdominal cavity environment.In this thesis,we propose an improved SLAM system that obtains a dense point cloud map by point cloud stitching,transforms the point cloud into a surface in the form of a triangular mesh using Poisson surface reconstruction,and superimposes the abdominal cavity texture image onto the mesh to generate a smooth inner wall texture on the surface.In this thesis,Hamlyn medical imaging databases are used as experimental data to design an evaluation experiment.The experimental results show that the system designed in this thesis improves the accuracy of feature point alignment compared to classical SLAM systems,while increasing the densities and providing more realistic reconstructed visual effects.