A method and device for generating a depth image

Abstract

The present invention is a method and device for generating a depth image, comprising: calculating the gray gradient of each scene pixel in a scene image block, and taking multiple scene pixels exceeding a gradient threshold as effective scene pixels. Obtain the initial coordinates and gray value of the sub-pixel corresponding to the same name in the reference image corresponding to the valid scene pixel, and express the parallax of the valid scene pixel as a linear polynomial function of the pixel coordinates. Obtain the normalized cross-correlation function value between the gray value of a valid scene pixel and the gray value of a sub-pixel with the same name. Get the target polynomial coefficient corresponding to the maximum value of the normalized cross-correlation function. Get the target disparity of the scene pixel at the center of the image patch. The depth value corresponding to the scene pixel is calculated from the disparity to generate the depth image corresponding to the scene image, and the accuracy of the depth estimation result of each pixel in the scene image is improved.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a method and device for generating a depth image. Background technique [0002] The monocular speckle structured light system is a depth perception device, which is used in intelligent security, mobile robots, augmented reality and other fields. The system uses a speckle projection device to project fine speckle patterns onto the surface of the scene, and at the same time, a camera captures a scene image, and matches the image with a pre-stored reference image to obtain a depth map of the scene. In this process, image matching is an important step to generate a depth map, and the quality of the matching result directly affects the accuracy of the depth result. [0003] Existing matching methods are mainly divided into two categories: grayscale-based methods and binary-based methods. These two types of methods generally use the idea of ​​image block matching to calculate ...

AI Technical Summary

Problems solved by technology

For more complex scene images, the parallax of each pixel in the local area may have a large range of variation, resulting in low accuracy of the depth estimation results for each pixel in the scene image.

Images