Point cloud reconstruction method and system based on auto-encoder

Abstract

The invention relates to a point cloud reconstruction method and system based on an auto-encoder. The method comprises the following steps: acquiring current data of a laser radar system; performing space division on the current data, and determining a plurality of data blocks; determining a local distance function field by adopting the trained auto-encoder network according to the data blocks; according to the local distance function field, using a contour surface extraction algorithm to determine a local scene surface; and splicing the local scene surfaces to determine the scene surface. According to the method, a high-quality complete scene can be reconstructed incrementally by a plurality of continuous frames of point clouds acquired by the laser radar, meanwhile, the scene storage overhead is relatively low, the reconstruction speed is relatively high, and the method can cope with the reconstruction work of most outdoor scenes.

Description

technical field [0001] The present invention relates to the fields of computer vision and computer graphics, in particular to a method and system for point cloud reconstruction based on an autoencoder. Background technique [0002] 3D reconstruction refers to the process of generating a 3D model from 2D image data or 3D point cloud data. 3D reconstruction has very important applications in computer vision, computer graphics, and robotics, and is the basis for applications such as autonomous driving, terrain generation, and augmented reality. [0003] 3D reconstruction is mainly divided into two categories, one is 3D reconstruction based on camera images, and the other is 3D reconstruction based on lidar. Image-based 3D reconstruction mainly includes camera calibration, feature point extraction, feature point matching to calculate pose, dense matching to estimate depth, and surface reconstruction. However, image-based 3D reconstruction is more dependent on texture and illum...

AI Technical Summary

Problems solved by technology

The point cloud reconstruction algorithm is mainly divided into two categories. One is the traditional reconstruction algorithm, which includes using the iterative nearest neighbor algorithm to solve the pose and using the voxel fusion algorithm to reconstruct the surface. This method can achieve incremental 3D reconstruction, but for large-scale In the scene, when updating the map, the storage overhead increases significantly

The second is an algorithm based on deep learning, which can use shape coding to store distance function fields, but the existing network needs to train a specific network for a specific object, which does not have generalization and cannot handle incremental point cloud data.

Images