基于多尺度注意力网络的立体匹配方法

doi:10.13190/j.jbupt.2020-204

北京邮电大学学报 ›› 2021, Vol. 44 ›› Issue (3): 27-34.doi: 10.13190/j.jbupt.2020-204

基于多尺度注意力网络的立体匹配方法

边继龙¹, 王厚博¹, 李金凤²

1. 东北林业大学信息与计算机工程学院, 哈尔滨 150040;
2. 牡丹江师范学院计算机与信息技术学院, 牡丹江 157011

收稿日期:2020-10-19 出版日期:2021-06-28 发布日期:2021-06-23
作者简介:边继龙(1982-),男,讲师,E-mail:bianjilong@nefu.edu.cn.
基金资助:
黑龙江省自然科学基金项目（F2018002）；黑龙江省基本科研业务费项目（1451MSYYB001）；中央高校基本科研业务费专项项目（2572016BB12）

Stereo Matching Method Based on Multiscale Attention Network

BIAN Ji-long¹, WANG Hou-bo¹, LI Jin-feng²

1. College of Information & Computer Engineering, Northeast Forestry University, Harbin 150040, China;
2. College of Computer & Information Technology, Mudanjiang Normal University, Mudanjiang 157011, China

Received:2020-10-19 Online:2021-06-28 Published:2021-06-23

摘要/Abstract

摘要： 为解决网络深度与训练图像块大小耦合问题及进一步提高弱纹理区域及边缘处的匹配精度，提出了一种基于多尺度注意力网络的立体匹配方法.该方法将立体匹配过程分为2个阶段：第1阶段提出了一种成本网络用于计算匹配成本，该网络由基础网络层和缩放层组成.第2阶段提出了一种基于多尺度注意力的视差求精网络，该视差求精网络综合了多种视差线索，并加入多尺度注意力机制进一步提高立体匹配精度.该方法在KITTI 2012、KITTI 2015和SceneFlow数据集上的3像素坏点百分比分别为1.13%，1.87%和2.29%.实验结果表明，与国内外同类方法相比，采用多尺度注意力网络的立体匹配方法在匹配精度上获得了较大的提升，尤其是在弱纹理区域及物体边缘处表现较好.

关键词: 深度网络, 立体匹配, 匹配成本, 多尺度注意力, 视差求精

Abstract: Aiming at the problem that the depth of network is related to the size of training image patches and improving the matching accuracy for the weak texture and edge regions, a multiscale attention network for stereo matching is presented. The method is divided into two stages:in the first stage, a deep network for computing matching cost is proposed, which is composed of basic layer and scale layer, in the second stage, a disparity refinement network based on multiscale attention is proposed, in which multiple disparity clues are combined and multiscale attention is added to boost stereo matching accuracy. The percentage of 3-pixel bad points on KITTI 2012, KITTI 2015, and SceneFlow is 1.13%, 1.87%, and 2.29%, respectively. Experiments show that compared with the current domestic and foreign advanced methods, a stereo matching method based on multiscale attention network made a great improvement in matching accuracy, especially better improvement for the weak texture and edge regions.

Key words: deep network, stereo matching, matching cost, multiscale attention, disparity refinement

中图分类号:

TP391.41

边继龙, 王厚博, 李金凤. 基于多尺度注意力网络的立体匹配方法[J]. 北京邮电大学学报, 2021, 44(3): 27-34.

BIAN Ji-long, WANG Hou-bo, LI Jin-feng. Stereo Matching Method Based on Multiscale Attention Network[J]. Journal of Beijing University of Posts and Telecommunications, 2021, 44(3): 27-34.

参考文献

[1] Bontar J, Lecun Y. Stereo matching by training a convolutional neural network to compare image patches[J]. Journal of Machine Learning Research, 2015, 17(1):2287-2318.
[2] Chen Zhuoyuan, Sun Xun, Wang Liang. A deep visual correspondence embedding model for stereo matching costs[C]//IEEE International Conference on Computer Vision. New York:IEEE, 2015:972-980.
[3] Luo Wenjie, Schwing A G, Urtasun R. Efficient deep learning for stereo matching[C]//IEEE Conference on Computer Vision and Pattern Recognition. New York:IEEE, 2016:5695-5703.
[4] He Kaiming, Zhang Xiangyu, Ren Shaoqing, et al. Deep residual learning for image recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition. New York:IEEE, 2016:770-778.
[5] Huang Gao, Liu Zhuang, Maaten L V D. Densely connected convolutional networks[C]//IEEE Conference on Computer Vision and Pattern Recognition. New York:IEEE, 2017:2261-2269.
[6] Chang Jiaren, Chen Yongsheng. Pyramid stereo matching network[C]//IEEE Conference on Computer Vision and Pattern Recognition. New York:IEEE, 2018:5410-5418.
[7] Guo Xiaoyang, Yang Kai, Yang Wukui, et al. Group-wise correlation stereo network[C]//IEEE Conference on Computer Vision and Pattern Recognition. New York:IEEE, 2019:3268-3277.
[8] Zhang Ke, Lu Jiangbo, Lafruit G. Cross-based local ste-reo matching using orthogonal integral images[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2009, 19(7):1073-1079.
[9] Hirschmuller H. Stereo processing by semiglobal matching and mutual information[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2008, 30(2):328-341.
[10] Fu Jun, Liu Jing, Tian Haijie, et al. Dual attention network for scene segmentation[C]//IEEE Conference on Computer Vision and Pattern Recognition. New York:IEEE, 2019:3141-3149.
[11] Yin Minghao, Yao Zhuliang, Cao Yue, et al. Disentangled non-local neural networks[C]//European Conference on Computer Vision. Cham:Springer, 2020:5000-5010.
[12] Cao Yue, Xu Jiarui, Lin Stephen, et al. GCNet:non-local networks meet squeeze-excitation networks and beyond[C]//IEEE International Conference on Computer Vision Workshop. New York:IEEE, 2019:1971-1980.
[13] He Kaiming, Zhang Xiangyu, Ren Shaoqing, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(9):1904-1916.
[14] Shaked A, Wolf L. Improved stereo matching with constant highway networks and reflective confidence learning[C]//IEEE Conference on Computer Vision and Pattern Recognition. New York:IEEE, 2017:6901-6910.
[15] 肖进胜, 田红, 邹文涛, 等. 基于深度卷积神经网络的双目立体视觉匹配算法[J]. 光学学报, 2018, 38(8):179-185. Xiao Jinsheng, Tian Hong, Zou Wentao, et al. Stereo matching based on convolutional neural network[J]. Acta Optica Sinica, 2018, 38(8):179-185.

基于多尺度注意力网络的立体匹配方法

Stereo Matching Method Based on Multiscale Attention Network

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

编辑推荐

Metrics

本文评价

[1]	韩晶张天鹏吕学强. 基于多模态特征与增强对齐的细粒度图像分类[J]. 北京邮电大学学报, 2024, 47(4): 130-135.
[2]	李云红朱景坤刘杏瑞陈锦妮苏雪平. 基于改进生成对抗网络的动漫图像风格迁移算法[J]. 北京邮电大学学报, 2024, 47(4): 117-123.
[3]	田昊王小玉. 融合暗通道先验与粒子群算法的去雾改进算法[J]. 北京邮电大学学报, 2024, 47(2): 118-122.
[4]	叶洪滨林政宽程红举. 基于多相机特征融合的行人检测算法[J]. 北京邮电大学学报, 2023, 46(5): 66-71.
[5]	常霄, 黄智濒, 禹旻, 杨武兵. 一种适用于航天大数据的深度决策树算法[J]. 北京邮电大学学报, 2023, 46(3): 1-6.
[6]	胡敏熊思黄宏程张光华王春芳. 基于RAPNet的脑肿瘤MRI图像三维分割[J]. 北京邮电大学学报, 2023, 46(2): 91-97.
[7]	荣震宇刘建毅. 基于Transformer和MLP的眼底血管分割算法[J]. 北京邮电大学学报, 2023, 46(1): 26-31.
[8]	张小乾谈振王潇梁芹万黎明. 基于正交投影学习的图像特征提取算法[J]. 北京邮电大学学报, 2022, 45(5): 85-90,128.
[9]	叶康张淑军郭淇李辉崔雪红. 基于CM-Transformer的连续手语识别[J]. 北京邮电大学学报, 2022, 45(5): 49-53,78.
[10]	王辰魁, 陈岳林, 蔡晓东. 图像样式风格迁移的行人再识别方法[J]. 北京邮电大学学报, 2021, 44(3): 67-72.
[11]	王睿男, 武穆清, 陈铁英, 李志刚, 孙建. 基于形态学的红外图像边缘检测[J]. 北京邮电大学学报, 2021, 44(1): 66-71.
[12]	熊光正, 黄智濒, 戴志涛, 杨武兵. 一种数据驱动的三维流场流线特征化筛选方法[J]. 北京邮电大学学报, 2019, 42(6): 91-97.
[13]	赵珊, 李永思. 基于主成分分析的图像哈希算法[J]. 北京邮电大学学报, 2019, 42(2): 36-41.
[14]	李喆, 李建增, 王哲. 基于先验优化的一致性模糊盲复原算法[J]. 北京邮电大学学报, 2019, 42(2): 63-69.
[15]	邹国锋, 傅桂霞, 王科俊, 高明亮, 申晋. 自适应深度卷积神经网络模型构建方法[J]. 北京邮电大学学报, 2017, 40(4): 98-103.