Real-time plane detection and extraction method based on depth image

Abstract

The invention discloses a real-time plane detection and extraction method based on a depth image, and the method comprises the steps: firstly carrying out the preprocessing of the depth image, extracting the block information of the depth image, and then carrying out the dimension reduction statistics of the normal vector distribution of a block; respectively detecting a parallel plane for each estimated normal vector to obtain the distance from each plane in each direction to the original point; and finally, clustering by taking the block as a clustering unit and the parameters of each plane as clustering seed points, performing plane parameter optimization and plane region extraction, and recording an ith plane parameter as Pi, the plane parameter including a normal vector of the plane and a distance between the plane and an original point. According to the method, the geometric information of all planes in the scene can be extracted from the depth image according to the imaging principle of the depth image and the spatial characteristics of the plane structure, so that a computer can acquire the plane structure information in the environment through the depth image in real time, and the three-dimensional reconstruction result is more accurate.

Description

technical field [0001] The invention relates to the fields of robot positioning and three-dimensional reconstruction, in particular to a method for detecting and extracting plane structures of depth images. Background technique [0002] As early as 2010, Microsoft (Microsoft) developed a low-cost structured light depth camera, which can obtain accurate depth images in a small indoor scene, thereby being able to reconstruct the indoor scene. At the 2011 SIGGRAPH (Special Interest Group on Graphics and Interactive Techniques, the abbreviation of Graphics and Interactive Technology Special Enthusiast Group) conference, Microsoft demonstrated the KinectFusion real-time reconstruction algorithm. Newcombe is the main person in charge of the algorithm project. Algorithm is the first algorithm to achieve real-time dense scene reconstruction. Based on this, Izadi completed the dynamic interaction between scenes and people. [0003] In 2016, Whelan proposed a method of detecting map...

AI Technical Summary

Problems solved by technology

The "Fast Plane Extraction in Organized Point Clouds Using Agglomerative Hierarchical Clustering" method [1] proposed by Feng et al. in 2014 can use 30 milliseconds to detect and extract the plane of a 640×480 resolution image, but 30 milliseconds is not enough for a real-time reconstruction system. It also takes quite a long calculation time. For example, the well-known 3D reconstruction work InfiniTAM-V3 takes about 5 milliseconds to reconstruct a frame of data, which shows that the current plane detection and extraction methods are not efficient enough.

Images