基于卷积神经网络的深度线段分类算法

doi:10.6040/j.issn.1672-3961.0.2019.416

摘要/Abstract

摘要： 为解决单目图像中冗余像素点不利于深度神经网络快速完成深度信息检测的问题,提出一种基于卷积神经网络的深度线段分类算法。对NYU-Depth数据集使用线段检测算法进行线段检测得到原始图像的线段特征图,通过数据预处理结合深度数据得到表征深度信息的线段集合及其标签,提出适用于线段特征的卷积神经网络,实现单目图像中深度线段的分类。通过在不同线段数目上进行多次多组对比试验,深度线段分类准确率达到73.50%。试验结果证明了利用卷积神经网络实现深度线段分类的可实施性,有助于更好的利用图像几何特征解决深度估计问题。

关键词: 单目图像, 深度估计, 卷积网络, 深度线段, 分类算法

Abstract: In order to solve the problem that redundant pixels in monocular images influenced depth information detection, a depth segment classification algorithm based on the convolutional neural network was proposed. We used NYU-Depth dataset to detect the segment-based features. Afterward, depth information was represented by line segments and its labels by the data preprocessing. The convolutional neural network was designed for considering the characteristics of the segments, and the classification of depth segments in monocular images was realized. By conducting several multi-group comparison experiments on different hyper-parameters, the accuracy of depth segment classification reached 73.50%. This experimental results proved the implement ability of the depth segment classification based on convolutional neural network, which was helpful to deep estimation using geometric features of images.

Key words: monocular image, depth estimation, convolutional network, depth segment, classification algorithm

中图分类号:

TP183

赵宁宁,唐雪嵩,赵鸣博. 基于卷积神经网络的深度线段分类算法[J]. 山东大学学报 (工学版), 2020, 50(4): 22-27.

ZHAO Ningning, TANG Xuesong, ZHAO Mingbo. Depth segment classification algorithm based on convolutional neural network[J]. Journal of Shandong University(Engineering Science), 2020, 50(4): 22-27.

参考文献

[1] ZHONG Y F, FFENG F, ZHANG L P. Large patch convolutional neural networks for the scene classification of high spatial resolution imagery[J]. Journal of Applied Remote Sensing, 2016, 10(2): 025006.
[2] MASSON G, YANG D, et al. Version and vergence eye movements in humans: open-loop dynamics determined by monocular rather than binocular image speed[J]. Vision Research, 2002, 42(26): 2853-2867.
[3] SHOTTON J, FITZGIBBON A, COOK M, et al. Real-time human pose recognition in parts from single depth images[J]. Communications of the ACM, 2013, 56(1): 116-124.
[4] KANEKO A M, YAMAMOTO K. Two-view monocular depth estimation by optic-flow-weighted fusion[J]. IEEE Robotics and Automation Letters, 2019, 4(2): 830-837.
[5] AURISANO A, RADOVIC A, ROCCO D, et al. A convolutional neural network neutrino event classifier[J]. Journal of Instrumentation, 2016, 11(9): P09001.
[6] LIU F, SHEN C, LIN G. Deep convolutional neural fields for depth estimation from a single image[C] //Computer Vision and Pattern Recognition(CVPR). Boston, USA: IEEE Computer Society, 2015: 5162-5170.
[7] BO L, DAI Y, HE M. Monocular depth estimation with hierarchical fusion of dilated CNNs and soft-weighted-sum inference[J]. Pattern Recognition, 2017, 83: 328-339.
[8] GRIGOREV A, JIANG F, RHO S, et al. Depth estimation from single monocular images using deep hybrid network[J]. Multimedia Tools and Applications, 2017, 76(18): 18585-18604.
[9] DAN X, RICCI E, OUYANG W, et al. Monocular depth estimation using multi-scale continuous CRFs as sequential deep networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2019, 41(6): 1426-1440.
[10] LI J, KLEIN R, YAO A, et al. A two-streamed network for estimating fine-scaled depth maps from single RGB images[J]. Computer Vision and Image Understanding, 2019, 186:25-36.
[11] WANG X, HOU C, PU L, et al. A depth estimating method from a single image using FoE CRF[J]. Multimedia Tools & Applications, 2015, 74(21): 9491-9506.
[12] XU H, JIANG M, LI F. Depth estimation algorithm based on data-driven approach and depth cues for stereo conversion in three-dimensional displays[J]. Optical Engineering, 2016, 55(12): 12106-1-12106-11.
[13] LI B, DAI Y, HE M. Monocular depth estimation with hierarchical fusion of dilated CNNs and soft-weighted-sum inference[J]. Pattern Recognition, 2018, 83: 328-339.
[14] CHEN L, PAPANDREOU G, KOKKINOS I, et al. Semantic image segmentation with deep convolutional nets and fully connected CRFs[J]. Computer Science, 2018, 40(4): 357-361.
[15] CAO Y, SHEN C, SHEN H. Exploiting depth from single monocular images for object detection and semantic segmentation[J]. IEEE Transactions on Image Processing, 2017, 26(2): 836-846.
[16] CAO Y, WU Z, SHEN C.Estimating depth from monocular images as classification using deep fully convolutional residual Networks[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2016, 38(10): 1-11.
[17] CHOI S, MIN D, HAM B, et al. Depth analogy: data-driven approach for single image depth estimation using gradient samples[J]. IEEE Transactions on Image Processing, 2015, 24(12): 5953-5966.
[18] QIN H, LI X, WANG Y, et al. Depth estimation by parameter transfer with a lightweight model for single still images[J]. IEEE Transactions on Circuits & Systems for Video Technology, 2017, 27(4): 748-759.
[19] 覃勋辉,马戎.一种基于梯度的直线段检测算法[J].光子学报, 2012,41(2):205-209. RONG X, MA R. A line segments detection algorithm based on grad[J]. Photonics Journal, 2012, 41(2): 205-209.
[20] GIOI R, MOREL J,et al. LSD: a fast line segment detector with a false detection control[J]. IEEE Transactions on Software Engineering, 2010, 32(4): 722-732.
[21] PATON K. Line detection by local methods[J]. Computer Graphics and Image Processing, 1979, 9(4): 316-332.
[22] BACON J, KING-SMITH P E. The detection of line segments[J]. Perception, 1977, 6(2): 125-131.

多维度评价

Viewed

Full text

Abstract

Cited

Shared

Discussed