A Depth Map Super-resolution Reconstruction Method Based on l1-l2 Penalty Function

Abstract

The invention discloses a depth map super-resolution reconstruction method based on L1-L2 penalty functions. The method comprises the following steps: a first step, calculating an initial estimated depth: mapping a low-resolution depth map to a high-resolution color image coordinate plane; a step 2, calculating a mask matrix: constructing a diagonal matrix, if the ith pixel point is located on a mapping coordinate of the low-resolution depth map on a high-resolution color image, the ith diagonal element is 1, and otherwise, the ith diagonal element is 0; a third step, calculating a neighbor matrix: constructing the neighbor matrix by means of the color similarity of pixel points of the high-resolution color image on a YCbCr space; a fourth step, constructing a depth super-resolution reconstruction model: establishing an energy model of depth super-resolution reconstruction; and a fifth step, resolving the energy model to obtain a reconstructed high-resolution depth map. The depth map super-resolution reconstruction method disclosed by the invention has the advantages that an object edge of the depth map can be enhanced while keeping the smooth transition of the depth in an object.

Description

technical field [0001] The present invention relates to the fields of computer vision and image processing, in particular to a depth map super-resolution reconstruction method based on an L1-L2 penalty function, which is used for stereo vision or super-resolution enhancement of images acquired by a depth sensor. Background technique [0002] Robot navigation, robot autonomous grasping, gesture recognition, collision detection, 3D reconstruction, industrial automation, and virtual reality all rely on high-resolution depth information. Although with the widespread popularity of depth cameras, we can obtain the depth of 3D scenes in real time; however, due to the limitation of sensors, these cameras can only obtain low-resolution depth maps. This limitation restricts the performance of depth cameras in the aforementioned applications. To this end, various methods have been proposed for the resolution enhancement of depth maps; depth map super-resolution reconstruction refers t...

AI Technical Summary

Problems solved by technology

The problem with existing methods is that it is difficult to maintain the object edges of the depth map, and the object edges contain the main energy of the depth map, which determines the effect of its subsequent application

Images