Real-time three-dimensional reconstruction method and system for large-scale scene based on line-of-sight updating algorithm

Abstract

The invention discloses a real-time three-dimensional reconstruction method and system for a large-scale scene based on a line-of-sight updating algorithm, and belongs to the field of computer visionand robots. The invention aims to solve a problem of providing a method in which a light of sight is generated according to the current three-dimensional point cloud, data updating is realized and thus real-time reconstruction is performed in allusion to a problem that general algorithms cannot realize real-time reconstruction for large-scale point cloud data of laser radar. The real-time three-dimensional reconstruction method mainly comprises the steps of acquiring three-dimensional point cloud, calculating external parameter data of a sensor, calculating a symbol distance value of an implicit surface based on a line-of-sight algorithm, performing weighted fusion on the symbol distance value, performing volume rendering on volume data, storing the volume data and displaying a reconstruction effect in real time. The implicit surface is updated in real time through introducing the line-of-sight updating algorithm, so that real-time reconstruction for a large-scale scene based on depthsensors such as radar is achieved. The real-time three-dimensional reconstruction method and system not only have great advantages in speed, but also acquire good effects in reconstruction quality.

Description

technical field [0001] The invention belongs to the field of computer vision and robots, and relates to a real-time three-dimensional reconstruction method and system for a large-scale scene based on a line-of-sight update algorithm. Background technique [0002] Real-time 3D reconstruction refers to the process of processing 3D point cloud or image data in real time and generating a 3D model. 3D reconstruction plays a very important role in the fields of computer vision and robotics, and is the basis for applications such as virtual reality, intelligent monitoring, and robot path planning. Since lidar can collect hundreds of thousands of points per second, the amount of data is huge, and the collected data contains relatively large noise. It is very challenging to reconstruct a 3D model of a large-scale scene in real time. [0003] 3D reconstruction is mainly divided into two categories: 3D reconstruction based on camera images and 3D reconstruction based on lidar. Image-...

AI Technical Summary

Problems solved by technology

Image-based 3D reconstruction has poor accuracy and robustness in feature point matching, and cannot achieve real-time 3D reconstruction

[0004] At present, the reconstruction algorithm based on lidar mainly adopts the method of simplifying the point cloud and then triangulating it. It is difficult to achieve incremental update of the surface model, and it is impossible to achieve real-time processing of each frame of 3D point cloud data.

In addition, the reconstruction quality is low due to the underutilization of the pose information of the sensor in each frame

Images