A Scene Reconstruction Method Based on Manhattan Hypothesis

Abstract

The invention discloses a scene reconstruction method based on Manhattan hypothesis. The method gives precise motion estimation based on a scheme of Manhattan hypothesis; firstly, a normal direction of all 3D points of a record frame is estimated; and then a normal vector direction of three main orthogonal planes is estimated via PCA (Principal Components Analysis). Normal vectors of the planes are estimated via all depth points, and random noise is greatly filtered, so that the estimated normal vectors are quite accurate; main plane coordinates are further determined via the position of each pixel in three-dimensional coordinates; and finally, the posture of a camera is estimated via a transformation matrix acquired from the plane coordinates and the normal vectors, and the camera postures of all frames of the image are spliced into a three-dimensional model of a scene. The plane information is obtained by calculating a large amount of points, so the method has more robustness than a single feature point method.

Description

technical field [0001] The invention belongs to the field of computer vision, and is especially aimed at three-dimensional scene reconstruction, and in particular relates to a scene reconstruction method based on the Manhattan assumption. Background technique [0002] In recent years, with the development of depth perception technology, it has become possible to realize real-time 3D scene scanning of 3D indoor scenes. Several systems have been proposed by the industry with promising results. On the other hand, as Augmented Reality (AR) becomes a hot topic in both academic and industry, real-time 3D scanning is urgently needed, as the restoration of the 3D geometry of our real scene is the key to seamless alignment of virtual objects. In Microsoft's Hololens head-mounted display, many AR-based applications need to scan the 3D geometry of the current room. [0003] Using a depth camera, which directly records 3D information, the key to achieving 3D scanning is to estimate th...

AI Technical Summary

Problems solved by technology

In this case, ICP may produce wrong camera motion

In addition, ICP requires a large number of sampling points and requires iterations to converge to the final correspondence, implying a relatively heavy computational cost

Even though some ICP-based systems use the GPU to achieve real-time performance, it is still not suitable for many practical applications, because the GPU may be occupied by other tasks, resulting in the ICP-based system not being able to calculate in time

Images