Color and depth information combined adaptive depth image restoration method

Abstract

The invention discloses a color and depth information combined adaptive depth image restoration method. The method comprises the steps of obtaining a to-be-restored depth image and a color original image before depth processing of the to-be-restored depth image; then, according to the mask of the to-be-restored depth map, obtaining cavity and noise point coordinates of the to-be-restored depth mapso as to confirm cavity and noise point coordinate information of the depth map, and carrying out image noise point restoration on noise points in the to-be-restored depth map by adopting a bilateralfiltering method; for holes in the to-be-repaired depth map, obtaining the to-be-repaired depth map; obtaining pixel points of which the pixel values cover the same positions in the depth map to be restored, wherein the pixel points are the same as the pixel points in the depth map to be restored in the color original image; the method achieves the restoration of the hole and noise information through employing a double-selection processing method, effectively solves a problem that a single filter cannot process a large-area hole in a depth image, reduces the restoration error of the depth image, achieves the reasonable restoration of the depth image of a scene, and achieves a very good effect and performance.

Description

technical field [0001] The invention relates to the field of depth image restoration of computer vision, in particular to an adaptive depth image restoration method combining color and depth information. Background technique [0002] Depth images play an important role in the field of computer vision, especially in 3D object detection. However, limited by the accuracy of the information collector and the complex image acquisition scene, the acquired depth map often has large areas of holes and noise. Take the indoor dataset NYUV2 as an example, such as figure 2 As shown, there are many large holes in the depth map of the NYUV2 dataset, and the image scene is complex, and there are phenomena such as stacking, large-area occlusion, and truncation between objects. For 3D object detection tasks, using depth maps to directly infer object 3D locations from 2D segmented regions suffers from large errors and uncertainties. Deep image inpainting has very important practical signif...

AI Technical Summary

Problems solved by technology

[0003] The purpose of the present invention is to provide an adaptive depth image repair method that combines color and depth information to overcome the problem of large image repair errors in existing depth image repair methods

Images