基于动态遍历的分层特征网络视觉定位

doi:10.19678/j.issn.1000-3428.0059101

计算机工程 ›› 2021, Vol. 47 ›› Issue (9): 197-202. doi: 10.19678/j.issn.1000-3428.0059101

基于动态遍历的分层特征网络视觉定位

蒋雪源, 陈青梅, 黄初华

贵州大学计算机科学与技术学院, 贵阳 550025

收稿日期:2020-07-30 修回日期:2020-09-03 发布日期:2021-09-13
作者简介:蒋雪源(1993-),男,硕士研究生,主研方向为计算机视觉、深度学习;陈青梅,硕士研究生;黄初华,副教授、博士。
基金资助:
贵州省自然科学基金（黔科合基础［2019］1088）；贵州大学引进人才科研项目（贵大人基合字（2017）31号，贵大人基合字（2015）52号）；贵州省教育厅创新群体重大研究项目（黔教合KY字［2018］026）。

Hierarchical Feature Network for Visual Localization Based on Dynamic Traversal

JIANG Xueyuan, CHEN Qingmei, HUANG Chuhua

College of Computer Science and Technology, Guizhou University, Guiyang 550025, China

Received:2020-07-30 Revised:2020-09-03 Published:2021-09-13

摘要/Abstract

摘要： 采用分层特征网络估计查询图像的相机位姿，会出现检索失败和检索速度慢的问题。对分层特征网络进行分析，提出采用动态遍历与预聚类的视觉定位方法。依据场景地图进行图像预聚类，利用图像全局描述符获得候选帧集合并动态遍历查询图像，利用图像局部特征描述符进行特征点匹配，通过PnP算法估计查询图像的相机位姿，由此构建基于MobileNetV3的分层特征网络，以准确提取全局描述符与局部特征点。在典型数据集上与AS、CSL、DenseVLAD、NetVLAD等主流视觉定位方法的对比结果表明，该方法能够改善光照与季节变化场景下对候选帧的检索效率，提升位姿估计精度和候选帧检索速度。

关键词: 视觉定位, 分层特征网络, 动态遍历, 预聚类, 位姿估计

Abstract: When used to estimate the camera pose of the query image, Hierarchical Feature Network(HFNet) is limited by frequent retrieval failures and the low retrieval speed.This paper analyzes HFNet and proposes a visual location method based on dynamic traversal and pre-clustering.According to the scene map, the image is pre-clustered.Then the global image descriptor is used to obtain the candidate frame set and dynamically traverse the query image, while the local feature descriptor is used to match the feature points. In addition, the camera pose of the query image is estimated by using the PnP algorithm.On this basis, an HFNet based on MobilenetV3 is constructed for the extraction of global descriptors and local feature points.Experimental results on typical data sets show that, compared with mainstream visual localization methods such as AS, CSL, DenseVLAD and NetVLAD, the proposed method can improve the retrieval efficiency of candidate frames in the cases of changing illumination conditions and seasons.It can also improve the accuracy of pose estimation and the speed of retrieving candidate frames.

Key words: visual localization, Hierarchical Feature Network(HFNet), dynamic traversal, pre-clustering, pose estimation

中图分类号:

TP391.41

蒋雪源, 陈青梅, 黄初华. 基于动态遍历的分层特征网络视觉定位[J]. 计算机工程, 2021, 47(9): 197-202.

JIANG Xueyuan, CHEN Qingmei, HUANG Chuhua. Hierarchical Feature Network for Visual Localization Based on Dynamic Traversal[J]. Computer Engineering, 2021, 47(9): 197-202.

http://www.ecice06.com/CN/Y2021/V47/I9/197

图/表 8

20210917191911

20210917191915

20210917191918

20210917191922

20210917191926

20210917191931

20210917191935

20210917191939

参考文献

[1] 刘浩敏, 章国锋, 鲍虎军.基于单目视觉的同时定位与地图构建方法综述[J].计算机辅助设计与图形学学报, 2016, 28(6):855-868. LIU H M, ZHANG G F, BAO H J.A survey of monocular simultaneous localization and mapping[J].Journal of Computer-Aided Design & Computer Graphics, 2016, 28(6):855-868.(in Chinese)
[2] HAN G J, JIANG J F, ZHANG C Y, et al.A survey on mobile anchor node assisted localization in wireless sensor networks[J].IEEE Communications Surveys & Tutorials, 2016, 18(3):2220-2243.
[3] 邓中亮, 尹露, 唐诗浩, 等.室内定位关键技术综述[J].导航定位与授时, 2018, 5(3):14-23. DENG Z L, YIN L, TANG S H, et al.A survey of key technology for indoor positioning[J].Navigation Positioning and Timing, 2018, 5(3):14-23.(in Chinese)
[4] SARLIN P E, DEBRAINE F, DYMCZYK M, et al.Leveraging deep visual descriptors for hierarchical efficient localization[C]//Proceedings of Conference on Robot Learning.Zurich, Switzerland:PMLR, 2018:456-465.
[5] SARLIN P E, CADENAC R, SIEGWART R, et al.From coarse to fine:robust hierarchical localization at large scale[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition.Washingtong D.C., USA:IEEE Press, 2019:12708-12717.
[6] LOWE D G.Distinctive image features from scale-invariant key points[J].International Journal of Computer Vision, 2004, 60(2):91-110.
[7] KIM H J, DUNN E, FRAHM J M.Learned contextual feature reweighting for image Geo-localization[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2017:3251-3260.
[8] ARANDJELOVIC R, GRONAT P, TORII A, et al.NetVLAD:CNN architecture for weakly supervised place recognition[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(6):1437-1451.
[9] SVARM L, ENQVIST O, KAHL F, et al.City-scale localization for cameras with known vertical direction[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(7):1455-1461.
[10] LIU L, LI H D, DAI Y C.Efficient global 2D-3D matching for camera localization in a large-scale 3D map[C]//Proceedings of IEEE International Conference on Computer Vision.Washington D.C., USA:IEEE Press, 2017:2391-2400.
[11] KENDALL A, GRIMES M, CIPOLLA R.PoseNet:a convolutional network for real-time 6-DOF camerare localization[C]//Proceedings of IEEE International Conference on Computer Vision.Washington D.C., USA:IEEE Press, 2015:2938-2946.
[12] SANDLER M, HOWARD A, ZHU M L, et al.MobileNetV2:inverted residuals and linear bottlenecks[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2018:4510-4520.
[13] DETONE D, MALISIEWICZ T, RABINOVICH A.SuperPoint:self-supervised interest point detection and description[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops.Washington D.C., USA:IEEE Press, 2018:337-350.
[14] HOWARD A, SANDLER M, CHEN B, et al.Searching for MobileNetV3[C]//Proceedings of IEEE/CVF International Conference on Computer Vision.Washington D.C., USA:IEEE Press, 2019:1314-1324.
[15] YANG T J, HOWARD A, CHEN B, et al.NetAdapt:platform-aware neural network adaptation for mobile applications[C]//Proceedings of European Conference on Computer Vision.Berlin, Germany:Springer, 2018:285-300.
[16] HU J, SHEN L, ALBANIE S, et al.Squeeze-and-excitation networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(8):2011-2023.
[17] CHOLLET F.Xception:deep learning with depth wise separable convolutions[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2017:1251-1258.
[18] CIPOLLA R, GAL Y, KENDALL A.Multi-task learning using uncertainty to weigh losses for scene geometry and semantics[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2018:7482-7491.
[19] FUKUI S, YU J, HASHIMOTO M.Distilling knowledge for non-neural networks[C]//Proceedings of NIPS'14.Washington D.C., USA:IEEEPress, 2019:1411-1416.
[20] SCHÖNBERGER J L, FRAHM J M.Structure-from-motionrevisited[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2016:4104-4113.
[21] SATTLER T, LEIBE B, KOBBELT L.Efficient & effective prioritized matching for large-scale image-based localization[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(9):1744-1756.
[22] TORII A, ARANDJELOVIĆ R, SIVIC J, et al.24/7 placere cognition by view synthesis[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2015:1808-1817.
[23] HU H J, WANG H S, LIU Z, et al.Retrieval-based localization based on domain-invariant feature learning under changing environments[C]//Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems.Washington D.C., USA:IEEE Press, 2019:3684-3689.

选择文件类型/文献管理软件名称

选择包含的内容

基于动态遍历的分层特征网络视觉定位

Hierarchical Feature Network for Visual Localization Based on Dynamic Traversal

RichHTML

PDF

可视化

被引次数

摘要/Abstract

引用本文

使用本文

图/表 8

参考文献

相关文章 6

编辑推荐

Metrics

本文评价

[1]	蒋明, 陈雨, 周青华, 袁媛, 何世琼. 适用于非合作目标捕获的轻量级位姿估计网络[J]. 计算机工程, 2022, 48(6): 235-242.
[2]	李少飞, 史泽林, 庄春刚. 基于深度学习的物体点云六维位姿估计方法[J]. 计算机工程, 2021, 47(8): 216-223.
[3]	马科伟, 张锲石, 康宇航, 任子良, 程俊. 移动机器人中视觉里程计技术综述[J]. 计算机工程, 2021, 47(11): 1-10.
[4]	赵德超,彭力,王皓. 非完整机器人目标跟踪控制器的设计与实现[J]. 计算机工程, 2019, 45(1): 297-302.
[5]	吕立,姚拓中,宋加涛,肖江剑,王建军. 基于单目视觉三维重建系统的设计与实现[J]. 计算机工程, 2018, 44(12): 233-239.
[6]	汪俊文;侯庭波;朱枫. 基于EIV模型的点线位姿估计[J]. 计算机工程, 2008, 34(6): 224-226.

模态框（Modal）标题

选择文件类型/文献管理软件名称

选择包含的内容

基于动态遍历的分层特征网络视觉定位

Hierarchical Feature Network for Visual Localization Based on Dynamic Traversal

RichHTML

PDF

可视化

被引次数

摘要/Abstract

引用本文

使用本文

图/表 8

参考文献

相关文章 6

编辑推荐

Metrics

本文评价