基于光照不变特征提取和动态特征祛除的视觉SLAM算法

doi:10.13474/j.cnki.11-2246.2025.0917

测绘通报 ›› 2025, Vol. 0 ›› Issue (9): 105-111.doi: 10.13474/j.cnki.11-2246.2025.0917

基于光照不变特征提取和动态特征祛除的视觉SLAM算法

柯学良¹, 肖玮¹, 曲乃铸¹, 何智杰^1,2, 黄睿¹

1. 联勤保障部队工程大学, 重庆 401331;
2. 31680部队, 四川崇州 611233

收稿日期:2025-03-14 发布日期:2025-09-29
通讯作者: 肖玮。E-mail:wzry@163.com
作者简介:柯学良(1997—),男,硕士生,主要研究方向为视觉SLAM。E-mail:kxl1997kxl@163.com
基金资助:
2023年重庆市教委科学技术研究重点项目(KJZD-K202312903);2024年重庆市研究生科研创新项目(CYS240831;CYS240832)

A visual SLAM algorithm based on illumination-robust feature extraction and dynamic feature removal

KE Xueliang¹, XIAO Wei¹, QU Naizhu¹, HE Zhijie^1,2, HUANG Rui¹

1. Joint Logistic Support Force Engineering University, Chongqing 401331, China;
2. Troops 31680, Chongzhou 611233, China

Received:2025-03-14 Published:2025-09-29

摘要/Abstract

摘要： 针对环境物体运动和光照变化导致视觉SLAM算法定位精度不高的问题,本文提出了一种适合动态变光环境的高精度视觉SLAM算法。该算法基于VINS Mono架构,首先,进行光照不变特征提取,即通过构造的ResNet网络,对输入图像进行特征提取,得到初始特征点坐标和光照不变特征图,随后利用光流法在光照不变特征图上对初始特征点进行光流估计,降低光照变化对特征跟踪的影响。然后,进行动态特征祛除,通过YOLOv8网络的像素级语义分割,将输入图像中的动态物体标记为掩膜,随后利用对极几何约束祛除掩膜范围内的动态特征,得到可稳定追踪的静态特征点,降低动态特征对算法定位精度的影响。最后,基于EuRoC、VIODE、Market数据集,将本文算法与VINS Mono算法进行对比试验,结果表明,本文算法的绝对轨迹误差在3组数据集上平均提升了55.09%,变光动态环境下得到了较好的定位精度。

关键词: 动态特征祛除, 光照不变特征提取, 光流法, VINS Mono

Abstract: To address the issue of low localization accuracy in visual SLAM algorithms caused by environmental moving objects and illumination changes,this paper proposes a high-precision visual SLAM algorithm suitable for dynamic and varying illumination environments.This algorithm is based on the VINS Mono architecture,firstly,it performs illumination-robust feature extraction by extracting features from input images through a ResNet network to obtain initial feature point coordinates and illumination-invariant feature maps; then,it performs optical flow estimation on these maps to reduce the impact of illumination on feature tracking.Subsequently,it carries out dynamic feature removal by using pixel-level semantic segmentation with YOLOv8 to mark dynamic objects in the input images as masks.Epipolar geometry constraints are then utilized to remove dynamic features within the mask regions,obtaining stable static feature points for tracking and reducing the impact of dynamic features on the algorithm's localization accuracy.Finally,comparative experiments on the EuRoC,VIODE,and Market datasets show that our method achieves 55.09%lower absolute trajectory error compared to VINS-Mono,demonstrating good localization accuracy in dynamic and varying illumination environments.

Key words: dynamic feature removal, illumination-robust feature extraction, optical flow method, VINS Mono

中图分类号:

P228

柯学良, 肖玮, 曲乃铸, 何智杰, 黄睿. 基于光照不变特征提取和动态特征祛除的视觉SLAM算法[J]. 测绘通报, 2025, 0(9): 105-111.

KE Xueliang, XIAO Wei, QU Naizhu, HE Zhijie, HUANG Rui. A visual SLAM algorithm based on illumination-robust feature extraction and dynamic feature removal[J]. Bulletin of Surveying and Mapping, 2025, 0(9): 105-111.

参考文献

[1] SINGANDHUPE A, LA H M.A review of SLAM techniques and security in autonomous driving[C]//Proceedings of the 3rd IEEE International Conference on Robotic Computing (IRC).Naples:IEEE, 2019: 602-607.
[2] WANG Yanan, TIAN Yaobin, CHEN Jiawei, et al.A survey of visual SLAM in dynamic environment: the evolution from geometric to semantic approaches[J].IEEE Transactions on Instrumentation and Measurement, 2024, 73:2523221.
[3] 万伟鹏.复杂光照和动态环境下的鲁棒视觉SLAM系统构建研究[D].北京:北京交通大学, 2023.
[4] 伍韬.基于前视声呐-视觉-惯性的水下SLAM算法研究[D].南京: 东南大学, 2023.
[5] 王坤鹏, 付世沫, 魏媛媛, 等.复杂光强下无人机视觉惯性定位算法[J].电光与控制, 2024, 31(11): 18-25.
[6] DETONE D, MALISIEWICZ T, RABINOVICH A.SuperPoint: self-supervised interest point detection and description[C]//Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW).Salt Lake City: IEEE, 2018: 337-33712.
[7] ZHAO Xiaoming, WU Xingming, MIAO Jinyu, et al.ALIKE: accurate and lightweight keypoint detection and descriptor extraction[J].IEEE Transactions on Multimedia, 2022, 25: 3101-3112.
[8] 刘亦博, 吴传文, 周宗锟, 等.融合SuperGlue方法的深度单目视觉惯性导航算法[J].测绘通报, 2024(2): 113-117.
[9] KLAPPSTEIN J, VAUDREY T, RABE C, et al.Moving object segmentation using optical flow and depth information[C]//Proceedings of 2009 Advances in Image and Video Technology.Berlin: Springer, 2009: 611-623.
[10] YU Chao, LIU Zuxin, LIU Xinjun, et al.DS-SLAM: a semantic visual SLAM towards dynamic environments[C]//Proceedings of 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).Madrid: IEEE, 2018: 1168-1174.
[11] YU Naigong, WANG Yang, MA Nan, et al.DRM-SLAM: robust visual SLAM based on dynamic region mask for dynamic scenes[C]//Proceedings of 2023 China Automation Congress (CAC).Chongqing: IEEE, 2023: 3811-3816.
[12] LIN Yicheng, WANG Shuo, JIANG Yunlong, et al.Breaking of brightness consistency in optical flow with a lightweight CNN network[J].IEEE Robotics and Automation Letters, 2024, 9(8): 6840-6847.
[13] WU Xiongwei, SAHOO D, HOI S C H.Recent advances in deep learning for object detection[J].Neurocomputing, 2020, 396: 39-64.
[14] BAKER S, SCHARSTEIN D, LEWIS J P, et al.A database and evaluation methodology for optical flow[J].International Journal of Computer Vision, 2011, 92(1): 1-31.
[15] WEI Tan, LIU Haomin, DONG Zilong, et al.Robust monocular SLAM in dynamic environments[C]//Proceedings of 2013 IEEE International Symposium on Mixed and Augmented Reality (ISMAR).Adelaide: IEEE, 2013: 209-218.

基于光照不变特征提取和动态特征祛除的视觉SLAM算法

A visual SLAM algorithm based on illumination-robust feature extraction and dynamic feature removal

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 2

编辑推荐

Metrics

本文评价

[1]	符强, 腾先云, 纪元法, 任风华, 孔健明. 基于双目深度筛选的ORB-SLAM3算法[J]. 测绘通报, 2024, 0(1): 77-82.
[2]	付振华, 邓瑞芝. 一种基于单目视觉的无人机仿地飞行方法[J]. 测绘通报, 2021, 0(11): 128-130.