Building three-dimensional point cloud registration method based on deep learning

Abstract

The invention discloses a building three-dimensional point cloud registration method based on deep learning. The method includes following steps: S1, obtaining training data; S2, constructing a deep learning model; S3, training, adjusting and optimizing the model; S4, preprocessing to-be-detected data; S5, detecting a point cloud key region; S6, detecting point cloud key points; S7, determining akey point corresponding relation; and S8, calculating a conversion relation and performing registration. According to the method, deep learning is applied to two key steps of point cloud registration:searching key points and determining a matching relation. According to the method, the key region and positioning key points are searched in steps for the representation capability of point cloud data by fully employing deep learning, compared with the conventional traversal mode, the detection speed is greatly accelerated, point cloud characteristics are learned by employing a deep network modelto replace manual characteristics, and the algorithm is more robust and efficient.

Description

technical field [0001] The invention relates to the field of three-dimensional urban reconstruction and unmanned driving, in particular to a method for registering three-dimensional point clouds of buildings based on deep learning. Background technique [0002] In industrial production, life and scientific research, there is a wide demand for 3D reconstruction, such as AR / VR, 3D printing, automation of industrial production, protection of cultural relics, and application of computer technology in the medical field. Nowadays, the accuracy of 3D scanning instruments is getting higher and higher, and it is becoming easier to obtain fine large-scale urban data. Some developed countries have already started the construction of smart cities, and the 3D reconstruction of cities is an important part of smart cities. For point cloud The research of data registration method becomes the key in the field of 3D reconstruction. [0003] In the traditional point-based registration method,...

AI Technical Summary

Problems solved by technology

This method cannot take into account all the characteristics of the point-based neighborhood, resulting in incomplete feature extraction. Secondly, the traversal method is computationally intensive, especially for the processing of point clouds with large-scale outdoor buildings.

The continuous and rapid development of deep learning at this stage has made major breakthroughs and fruitful scientific research results in many fields, especially in the field of image-related research. just started

Especially in the application of 3D point cloud, the point cloud is basically preprocessed into certain specifications by projecting and dividing into low dimensions, which will cause a large loss of information

Images