Boundary feature preserved depth image enhancement method

Abstract

The invention discloses a boundary feature preserved depth image enhancement method and belongs to the field of depth image processing. The implementation method includes: calibrating a color camera and a depth camera to acquire color-depth image group with pixels being in one-to-one correspondence; subjecting a depth image to mid-value filtering processing to acquire a depth image reliable in depth value; acquiring a cavity mask of the depth image reliable in the depth value and determining the boundary direction of a cavity by adopting Sobel gradient transformation; subjecting the acquired color image and cavity mask to directional combined bilateral filtering to acquire the enhanced depth image; calculating cavity pixel percentage HPP of the enhanced depth image; subjecting the depth image reliable in depth to interative filtering to acquire depth value of the cavity area of the depth image and filing the cavity area of the depth image. Through iterative condition judgment, denoising and cavity filling of the depth image are realized, and the final enhanced depth image capable of keeping the image quality clear and features complete is outputted.

Description

technical field [0001] The invention belongs to the field of depth image processing, in particular to a depth image enhancement method for preserving boundary features. Background technique [0002] In recent years, stereoscopic imaging technology has developed rapidly. The depth image containing scene depth information is an important clue for environment perception and scene understanding, and has been widely used in the field of computer vision, such as: robot navigation, human-computer interaction, 3D scene reconstruction, etc. At present, there are mainly two types of sensors for obtaining depth information: one is a depth sensor based on ToF technology, and the other is a depth sensor based on structured light technology. Depth sensors based on ToF technology measure spatial distance using the time-of-flight of pulsed light of a specific frequency in space. This technology has the disadvantages of low resolution of the acquired depth image, and the ambient light can s...

AI Technical Summary

Problems solved by technology

The current depth image enhancement method mainly has the defects of inaccurate filling of the depth value at the hole and poor object edge preservation characteristics. The similarity between the pixels of the color image corresponding to the image is used as the basis for depth diffusion, and the depth filling of the hole area is completed based on the known depth of the depth image

However, this method fills the same color objects at different positions in the space with the same depth value, resulting in inaccurate depth filling and blurred boundaries.

Changchun University of Science and Technology's CN106780383 patent "TOF camera depth image enhancement method" proposed to use filtering and interpolation methods to suppress depth image noise and improve depth image resolution, but did not solve the hole defects in depth images

Images