Research And Implementation Of Object 3D Reconstruction Method Based On Monocular Multi-view Stereo Geometry

3D reconstruction system aims to replicate the human visual system's understanding of space,using useful geometric and semantic information from 2D images obtained by the camera to restore 3D structure,information and motion state of objects in the environment.Realizing3 D reconstruction based on the information of 2D images has achieved great results in many fields,but factors such as robustness,accuracy,completeness,and efficiency are still key challenges for image three-dimensional reconstruction.This subject designs a 3D reconstruction algorithm with a monocular camera as the sensor,and uses object constraints to improve the reconstruction accuracy and efficiency.The features used in traditional 3D reconstruction algorithms are mainly low-level features in the image,such as feature points,lines,etc.It is easy to lack good adaptation and generalization in a single texture,areas with insignificant gradient changes,and complex changes.For the question of chemistry,a 3D object reconstruction model based on multi-view three-dimensional geometry is proposed.First,the multi-view images taken around the object are processed for image field of view expansion,and then the geometric method is used for 2D target detection,and the 2D object detection results are used to constrain the feature extraction and matching algorithms to limit the matching feature points to Reduce mismatching within a certain area and improve matching accuracy.The traditional multi-view 3D reconstruction algorithm Sf M is studied,and the classic BA optimization algorithm is discussed.First,use the vanishing point of the perspective principle and the 2D candidate frame to solve the 3D object information.In view of the uncertainty of the direction angle of the generated 3D object results,the candidate frame scoring function is designed to select the best 3D object detection results;For the BA algorithm optimization in traditional Sf M,only the constraints between the spatial points and the camera are used.The optimal 3D object detection data is used as a new constraint item to improve the 3D reconstruction BA optimization;Realize the pairwise constraint between the camera and the 3D point cloud and the object.The addition of 3D object information improves the accuracy of camera pose estimation and the accuracy of scene reconstruction in complex scenes.Complete the architecture design and software implementation of the algorithm,and implement the entire system in Ubuntu and ROS operating systems.Use TUM and self-collected data sets to test the feature extraction matching part and the 3D reconstruction part respectively to evaluate the matching accuracy and reconstruction accuracy,and verify the effectiveness of the algorithm and system proposed in this paper.