Research On 3D Modeling Method Of Internal Cavity Based On ORB-SLAM

In order to simplify the navigation system for minimally invasive laparoscopic surgery,a SLAM-based laparoscopic navigation system was produced.Fusion of medical image information with real-time laparoscopic video can greatly increase the field of vision of surgeons,enable them to see key site structures that are below the surface,and achieve safer and more effective surgery.However,due to the humid environment in the cavity,there will be a large number of specular reflections related to the camera position in the cavity image.Specular reflections are common problems that cause inaccurate depth estimation and map reconstruction.Therefore,this article studies how to apply the SLAM system to 3D reconstruction of the lumen.The structural parameters of the monocular system are mainly designed,and a feature matching algorithm suitable for the internal cavity is studied,and a dense 3D reconstruction of the extracted point cloud is performed.The thesis mainly includes the following work:Firstly,this article studies the basic principles of SLAM technology and analyzes the advantages and disadvantages of SLAM systems based on direct and indirect methods.Find the SLAM system that is most suitable for 3D reconstruction of the internal cavity.Since SLAM was originally developed for robot navigation and is not very specific to the internal cavity,this article has adjusted the parameters of the SLAM system suitable for the internal cavity to make it able to It is more suitable for model establishment in minimally invasive surgery.Secondly,due to the complexity and particularity of the internal cavity environment,organic surfaces with specular reflection and uniform texture,commonly used feature point extraction algorithms such as FAST,SURF,SHIFT,ORB can not well extract stable features.In order to find stable and repeatable features,this paper proposes a feature extraction method of branch points and branch segments based on lumen vessel characteristics.The performance evaluation and comparison of the methods were performed based on the medical image sequence.The experiments show that these features are more stable and effective.Finally,the point cloud obtained from the key frame image is subjected to least squares optimization and Poisson surface reconstruction to obtain dense surface reconstruction.Using public medical data sets,experiments are performed on the sign system and 3D reconstruction proposed in this paper.It can be seen that the improved internal cavity reconstruction system based on ORB-SLAM designed in this paper can well locate the endoscope and generate a high-precision point cloud,and finally generate a dense 3D reconstruction model with texture.