Research On 3D Point Cloud Reconstruction Algorithm Based On Monocular Vision Multi-View Geometry

How to reconstruct the world with 2D images is always a hot topic in the field of computer vision.There are many ways to reconstruct based on monocular vision.One of the methods is to study the relationship between different views through multi-view geometry under the premise of knowing the parameters in the camera,and use this relationship to complete reconstruction.This reconstruction method has extensive research value and application scenarios both in laboratory research and industrial landing due to the easy accessibility of its equipment and the convenience of data collection.In order to solve the problem of how to reconstruct the point cloud from a set of multi-view images of the target,an improved feature extraction and matching algorithm based on scale-invariant feature transform is proposed.Based on the feature matching algorithm based on Euclidean distance and ratio verification,The reverse verification of the cosine similarity further filters and purifies the matching result.In order to solve the problem that feature extraction based on scale-invariant feature transform has more interference points when reconstructing,an improved feature extraction algorithm based on deep learning is proposed.The two-dimensional point corresponding to the spatial point is used to train the network to increase the central target.The proportion of feature points in the area to improve the quality of reconstruction.On the basis of improved feature extraction and matching,the camera pose is restored by calculating the external geometric constraint relationship between the matching views,and then the reconstruction of the three-dimensional space is completed by combining the triangulation.Finally,the reconstruction result is processed,and the beam method is used to minimize the error to obtain accurate reconstruction results.To verify the performance of the algorithm,experiments were performed using the Oxford University Visual Group online 3D dataset.By comparing the feature extraction and matching results,the feature points are verified in the central target area ratio to improve the feature extraction and matching algorithm performance.The reconstruction quality is analyzed by the reconstruction error combined with the proportion of the feature points and the reconstruction points.Experiments show that the algorithm further reduces the number of mismatches and achieves better quality reconstruction results.