Monocular Depth Matching With Hybrid Sampling and Depth Label Propagation

Monocular Depth Matching With Hybrid Sampling and Depth Label Propagation

Ye Hua, Qu Xi Long, Li Zhen Jin
Copyright: © 2022 |Pages: 14
DOI: 10.4018/IJDCF.302879
Article PDF Download
Open access articles are freely available for download

Abstract

This paper proposes a monocular depth label propagation model, which describes monocular images into depth label distribution for the target classification matching. 1) Depth label propagation by hybrid sampling and salient region sifting, improve the discrimination of detection feature categories. 2) Depth label mapping and spectrum clustering to classify target, define the depth of the sorting rules. The experimental results of motion recognition and 3D point cloud processing, show that this method can approximately reach the performance of all previous monocular depth estimation methods. The neural network model black box training learning module is not used, which improves the interpretability of the proposed model.
Article Preview
Top

Depth estimation and 3D reconstruction of a single image are still very challenging now. Difficulty in identification of the motion of occluded targets can be decomposed into three sub-problems: occlusion target division and sparse depth estimation, overall image depth estimation and motion recognition. The following is the method proposed in this paper to solve the three sub propositions.

  • 1)

    Hybrid sampling of salient region

In section 3.1 sparse processing methods (Cai et al.,2011)(Ye et al.,2018)(Jiao et al.,2017)(Liu et al.,2013)(Wang et al.,2015)(Maaten et al.,2008) is applied to detect the salient region, using hybrid sampling for the image information in the scene space structure, calculating the ambiguity of the salient region.

Complete Article List

Search this Journal:
Reset
Volume 16: 1 Issue (2024)
Volume 15: 1 Issue (2023)
Volume 14: 3 Issues (2022)
Volume 13: 6 Issues (2021)
Volume 12: 4 Issues (2020)
Volume 11: 4 Issues (2019)
Volume 10: 4 Issues (2018)
Volume 9: 4 Issues (2017)
Volume 8: 4 Issues (2016)
Volume 7: 4 Issues (2015)
Volume 6: 4 Issues (2014)
Volume 5: 4 Issues (2013)
Volume 4: 4 Issues (2012)
Volume 3: 4 Issues (2011)
Volume 2: 4 Issues (2010)
Volume 1: 4 Issues (2009)
View Complete Journal Contents Listing