An optimization method for semantic segmentation of rgbd images based on depth density

Abstract

The invention discloses a depth density-based RGBD image semantic segmentation optimization method, which belongs to the field of computer image processing. It includes the following steps: calculating the average depth μ of pixels in the range of n×n centered on the (x, y) pixel in the RGBD image x，y : where, d x，y is the depth value of the (x, y) point on the image, and the image size is h×w; calculate the RGBD image with the (x, y) point as the center, within the range of n×n and the (x, y) point pixel Depth variance σ x，y : Calculate the average depth μ in the range of n × n with the (x, y) point as the center in the RGBD image x，y The depth variance of the image is added to the padding, and the depth value of the padding is 0, so that the image size becomes (h+(n‑1) / 2, w+(n‑1) / 2), and the image to be segmented is obtained; In the present invention, the depth image is used to calculate the depth density of each position in the picture, and the depth density is used to judge whether adjacent regions belong to the same object, so that the effect of semantic segmentation can be effectively improved.

Description

technical field [0001] The invention relates to the field of computer image processing, in particular to a depth density-based RGBD image semantic segmentation optimization method. Background technique [0002] RGBD is an image type, its essence is RGB+Depth, that is, in the process of image acquisition, the depth information of the target (the linear distance from the target surface to the lens) will be obtained at the same time, and the RGBD image in this patent uses ToF (Time of Fly) technology to obtain, this type of technology is characterized by fast imaging, high precision, can achieve real-time acquisition of two types of images. The disadvantage is that the resolution of the depth image is relatively low. [0003] The deep convolutional network is a key technical point in the field of deep learning. Its basis is a multi-layer neural network. The difference is that the full connection of the original neural network is converted into a convolution operation, which wi...

AI Technical Summary

Problems solved by technology

[0006] At present, when using a full convolutional network for image segmentation, the feature map (heat map) is restored to the original image size, and the segmentation result is too rough and the boundary is not clear

Images