Scene flow estimation method based on 3D local rigidity and depth map guided anisotropic smoothing

Abstract

The invention relates to a scene flow estimation method based on 3D local rigidity and depth map guided anisotropic smoothing. The method comprises the following steps: S1, acquiring a texture image and a depth image which are aligned at the same time through an RGB-D sensor; S2, building a scene flow estimation energy functional, and calcualting a dense scene flow based on a 3D local rigidity surface hypothesis and a global constrained method, wherein the form of a scene flow energy function is shown in the description; S3, designing date items based on the texture image and the depth image as well as the 3D local rigidity surface hypothesis; S4, designing smoothing items based on a depth map driven anisotropic diffusion tensor and total variation regularization; S5, creating an image pyramid, and adopting a coarse-to-fine solution strategy; and S6, calculating a scene flow by use of a duality method, and introducing scene flow auxiliary variables. According to the invention, the weight of a space-domain filter is determined by both the chromatic aberration and location relationship between the pixels of a color image, and therefore, the edge distortion problem in the process of repair is solved. In order to reduce repair error, the weight of a value-domain filter is determined by the color information and the structural similarity coefficient.

Description

technical field [0001] The present invention relates to a scene flow estimation method based on 3D local rigidity and depth map guided anisotropic smoothing. Background technique [0002] Scene flow is a 3D dense motion field that represents the 3D motion of each point in the actual scene. Scene flow represents the real motion field of the scene and contains depth information, so scene flow has broad application prospects in the fields of intelligent human-computer interaction, 3D reconstruction and vehicle assisted driving. Accurately estimating scene flow is a crucial problem for its widespread application. Scene flow solution is an ill-conditioned problem, which requires a variety of assumptions and constraints to solve. [0003] The concept of scene flow was first proposed by Vedula of Carnegie Mellon University in 1999. Scene flow estimation is mainly divided into: scene flow estimation based on binocular stereo vision and scene flow estimation based on depth sensor....

AI Technical Summary

Problems solved by technology

However, in the existing technology, few people pay attention to solving the accuracy of the scene flow and also pay attention to the preservation effect of the moving edge.

Images