TOHF：一种针对资源受限室内VSLAM的特征提取器

doi:10.16182/j.issn1004731x.joss.23-1334

摘要/Abstract

摘要：

为解决现有VSLAM特征提取器在室内环境中对纹理和光照变化敏感、特征点冗余导致的局部依赖性过强以及硬件资源受限时的存储开销问题，提出了一种面向纹理的均匀FAST特征提取器（texture-oriented and homogenized FAST feature extractor，TOHF）。结合HVS（human visual system），采用二阶段阈值策略来更敏感地应对纹理的清晰度和复杂度差异。根据特征点密度的变化来动态调整特征点的分布，在兼顾计算效率和存储开销的同时，保证特征点分布结构信息。在资源受限设备录制的数据集和官方EuRoc数据集上基于ORB-SLAM3框架开展实验，采用匹配率、重投影误差、绝对轨迹误差（ATE）和耗时作为评估指标。实验结果表明：TOHF在视觉加惯导模式下带来更高精度和鲁棒性的同时，仍满足实时性要求。

关键词: VSLAM, HVS, 特征提取, ORB-SLAM3, 纹理

Abstract:

To address the issues of sensitivity to texture and lighting variations, excessive local dependence caused by feature point redundancy, and storage overhead under hardware resource constraints in existing VSLAM feature extractors in indoor environments, We propose the Texture-Oriented and Homogenized FAST Feature Extractor (TOHF), which integrates HVS (Human Visual System) for enhanced texture analysis. TOHF employs a two-stage thresholding strategy and dynamically adjusts feature point distribution, balancing computational efficiency and storage needs. We conducted experimental verification based on the ORB-SLAM3 framework on dataset from resource-limited device and the EuRoc dataset, focusing on matching rate, reprojection error, absolute trajectory error(ATE), and time efficiency. Results show that TOHF improves accuracy and robustness in vision-inertial navigation modes while maintaining real-time performance.

Key words: VSLAM, human visual system(HVS), feature extract, ORB-SLAM3, texture

中图分类号:

TP391.9

李若晴,赵瑶池,胡祝华等 . TOHF：一种针对资源受限室内VSLAM的特征提取器[J]. 系统仿真学报, 2025, 37(3): 691-703.

Li Ruoqing,Zhao Yaochi,Hu Zhuhua,et al . TOHF: A Feature Extractor for Resource-constrained Indoor VSLAM[J]. Journal of System Simulation, 2025, 37(3): 691-703.

图/表 13

图1

图2

图3

图4

图5

表1

原图的特征匹配结果比较

方法	场景	$N k 1$	$N k 2$	$N m$	$P / %$
FAST (阈值7)	简单	2	0	0	0
	复杂	3 823	3 647	3 373	92.49
	一般	5 401	5 128	4 109	80.13
FAST (阈值20)	简单	0	0	0	0
	复杂	1 790	1 730	1 689	97.63
	一般	1 484	1 037	891	85.92
BRISK	简单	1	0	0	0
	复杂	1 271	1 206	1 108	91.87
	一般	1 874	1 734	1 508	86.97
AKAZE	简单	0	0	0	0
	复杂	554	525	489	93.14
	一般	1 128	1 061	956	89.16
TOHF	简单	86	60	50	83.33
	复杂	814	760	738	97.10
	一般	1 199	951	857	90.12

表1

表2

对比度降低60%亮度减少60的特征匹配结果

方法	场景	$N k 1$	$N k 2$	$N m$	$P / %$
FAST (阈值7)	简单	0	0	0	0
	复杂	733	639	621	97.18
	一般	1 137	954	843	88.36
FAST (阈值20)	简单	0	0	0	0
	复杂	102	94	76	80.85
	一般	274	248	211	85.08
BRISK	简单	0	0	0	0
	复杂	14	11	5	45.45
	一般	150	132	112	84.85
AKAZE	简单	0	0	0	0
	复杂	3	1	0	0
	一般	176	116	101	87.07
TOHF	简单	0	0	0	0
	复杂	685	616	598	97.08
	一般	100	84	78	92.86

表2

表3

对比度降低60%亮度增加80的特征匹配结果

方法	场景	$N k 1$	$N k 2$	$N m$	$P / %$
FAST (阈值7)	简单	0	0	0	0
	复杂	1417	1281	1200	93.68
	一般	1593	1281	1104	86.18
FAST (阈值20)	简单	0	0	0	0
	复杂	261	217	189	87.10
	一般	447	386	341	88.34
BRISK	简单	0	0	0	0
	复杂	156	134	116	86.57
	一般	523	516	455	88.18
AKAZE	简单	0	0	0	0
	复杂	106	89	78	87.64
	一般	434	417	371	88.97
TOHF	简单	71	66	54	81.82
	复杂	763	696	653	93.82
	一般	427	373	345	92.49

表3

图6

表4

TOHF的定量比较

方法	场景	$N k$	$N h$	$N m$	$P / %$
特征提取	简单	86	—	50	83.33
	复杂	814	—	638	96.67
	一般	1 199	—	407	90.24
四叉树方法	简单	86	42	37	88.09
	复杂	814	184	178	96.74
	一般	1 199	973	897	92.19
TOHF	简单	86	53	49	92.45
	复杂	814	283	278	98.23
	一般	1 199	287	264	91.99

表4

表5

EuRoc数据集不同子数据集上的实验结果

数据集	ORB-SLAM3 (20,7)			面向纹理的特征提取 (消融1)			密度感知均匀化处理 (消融2)			TOHF-SLAM
数据集	ATE	$N k f$	$t / s$	ATE	$N k f$	$t / s$	ATE	$N k f$	$t / s$	ATE	$N k f$	$t / s$
MH01	0.057	3 681	203.7	0.029	3 680	213.7	0.025	3 679	204.4	0.018	3 687	209.9
MH02	0.068	3 043	180.3	0.038	3 042	179.1	0.019	3 040	169.9	0.021	3 047	177.7
MH03	0.059	2 704	154.0	0.047	2 697	150.6	0.038	2 706	155.8	0.036	2 705	155.3
MH04	0.129	2 029	114.0	0.108	2 031	114.2	0.082	2 033	115.9	0.097	2 032	114.2
MH05	0.077	2 274	127.2	0.053	2 272	130.4	0.058	2 273	1276	0.053	2 271	129.6
V101	0.045	2 913	160.8	0.038	2 916	163.8	0.041	2 912	162.6	0.040	2 910	165.9
V102	0.027	1 705	97.1	0.014	1 704	97.1	0.014	1 710	99.2	0.013	1 716	98.6
V103	0.036	2 149	120.1	0.028	2 144	119.8	0.031	2 153	127.6	0.025	2 140	123.7
V201	0.051	2 283	128.1	0.043	2 276	132.4	0.042	2 281	127.6	0.041	2 287	128.5
V202	0.019	2 346	132.1	0.011	2 353	145.1	0.016	2 344	131.4	0.014	2 339	130.7
V203	0.053	1 994	128.8	0.018	1 992	128.5	0.030	1 999	132.6	0.021	1 987	127.9
平均值	0.056	—	140.6	0.039	—	143.1	0.036	—	141.3	0.035	—	142.0

表5

图7

图8

参考文献 41

1	Chen Weifeng, Shang Guangtao, Ji Aihong, et al. An Overview on Visual SLAM: From Tradition to Semantic[J]. Remote Sensing, 2022, 14(13): 3010.
2	王浩宇, 龚光红, 蔡继红, 等. 基于战场元宇宙的动态三维场景感知[J]. 系统仿真学报, 2023, 35(10): 2262-2278.
	Wang Haoyu, Gong Guanghong, Cai Jihong, et al. Dynamic 3D Scene Perception Based on Battlefield Metaverse[J]. Journal of System Simulation, 2023, 35(10): 2262-2278.
3	Tourani Ali, Bavle Hriday, Jose Luis Sanchez-Lopez, et al. Visual SLAM: What Are the Current Trends and What to Expect?[J]. Sensors, 2022, 22(23): 9297.
4	Yin Jun, Luo Dongting, Yan Fei, et al. A Novel Lidar-assisted Monocular Visual SLAM Framework for Mobile Robots in Outdoor Environments[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 1-11.
5	Wenzel Patrick, Yang Nan, Wang Rui, et al. 4Seasons: Benchmarking Visual SLAM and Long-term Localization for Autonomous Driving in Challenging Conditions[J]. International Journal of Computer Vision, 2024: 1-23.
6	Shen Xiuqiang, Chen Lihang, Hu Zhuhua, et al. A Closed-loop Detection Algorithm for Online Updating of Bag-of-words Model[C]//Proceedings of the 2023 9th International Conference on Computing and Data Engineering. New York: ACM, 2023: 34-40.
7	Chen Mingyou, Tang Yunchao, Zou Xiangjun, et al. 3D Global Mapping of Large-scale Unstructured Orchard Integrating Eye-in-hand Stereo Vision and SLAM[J]. Computers and Electronics in Agriculture, 2021, 187: 106237.
8	于雅楠, 史敦煌, 华春杰. 特征点法SLAM视觉里程计自适应优化算法[J]. 系统仿真学报, 2022, 34(1): 104-112.
	Yu Yanan, Shi Dunhuang, Hua Chunjie. Adaptive Optimization in Feature-based SLAM Visual Odometry[J]. Journal of System Simulation, 2022, 34(1): 104-112.
9	Fu Yuexin, Han Bing, Hu Zhuhua, et al. CBAM-SLAM: A Semantic SLAM Based on Attention Module in Dynamic Environment[C]//2022 6th Asian Conference on Artificial Intelligence Technology (ACAIT). Piscataway: IEEE, 2022: 1-6.
10	Qi Hao, Hu Zhuhua, Xiang Yunfeng, et al. ATY-SLAM: A Visual Semantic SLAM for Dynamic Indoor Environments[C]//Advanced Intelligent Computing Technology and Applications. Singapore: Springer Nature Singapore, 2023: 3-14.
11	Cai Dupeng, Hu Zhuhua, Li Ruoqing, et al. AGAM-SLAM: An Adaptive Dynamic Scene Semantic SLAM Method Based on GAM[C]//Advanced Intelligent Computing Technology and Applications. Singapore: Springer Nature Singapore, 2023: 27-39.
12	Sengupta A, Ye Yuting, Wang R, et al. Going Deeper in Spiking Neural Networks: VGG and Residual Architectures[J]. Frontiers in Neuroscience, 2019, 13: 95.
13	林士琪, 王纪凯, 裴浩渊, 等. 基于物体关系描述的单目语义SLAM方法[J]. 系统仿真学报, 2022, 34(2): 278-284.
	Lin Shiqi, Wang Jikai, Pei Haoyuan, et al. Monocular Semantic SLAM Method Based on Object Relation Description[J]. Journal of System Simulation, 2022, 34(2): 278-284.
14	Sun Jiaming, Shen Zehong, Wang Yuang, et al. LoFTR: Detector-free Local Feature Matching with Transformers[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2021: 8918-8927.
15	Lindenberger Philipp, Sarlin Paul-Edouard, Pollefeys Marc. LightGlue: Local Feature Matching at Light Speed[C]//2023 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2023: 17581-17592.
16	Shi Jianbo, Tomasi C. Good Features to Track[C]//1994 Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 1994: 593-600.
17	Harris C, Stephens M. A Combined Corner and Edge Detector[C]//Proceedings of the Alvey Vision Conference. Durham: Alvety Vision Club, 1988: 23.1-23.6.
18	Chen Xinyu, Yu Yantao. HLE-SLAM: SLAM for Overexposed Construction Environment[C]//ISARC. Proceedings of the International Symposium on Automation and Robotics in Construction. IAARC Publications, 2023, 40: 585-588.
19	Davison A J, Reid I D, Molton N D, et al. MonoSLAM: Real-time Single Camera SLAM[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2007, 29(6): 1052-1067.
20	Lowe David G. Distinctive Image Features from Scale-invariant Keypoints[J]. International Journal of Computer Vision, 2004, 60(2): 91-110.
21	Munich M E, Pirjanian P, Di Bernardo E, et al. SIFT-ing Through Features with ViPR[J]. IEEE Robotics & Automation Magazine, 2006, 13(3): 72-77.
22	Bay Herbert, Ess Andreas, Tuytelaars Tinne, et al. Speeded-up Robust Features (SURF)[J]. Computer Vision and Image Understanding, 2008, 110(3): 346-359.
23	Viswanathan D G. Features from Accelerated Segment Test (Fast)[C]//Proceedings of the 10th Workshop on Image Analysis for Multimedia Interactive Services. [S.l.: s.n.], 2009: 6-8.
24	Zhang Jie, Xiong Shuai, Liu Cheng, et al. FPGA-based Feature Extraction and Tracking Accelerator for Real-time Visual SLAM[J]. Sensors, 2023, 23(19): 8035.
25	Lei Hua, Zhang Yi, Zhao Xiang, et al. A Stereo Matching Algorithm for Fast Extraction of Binocular Depth in Visual SLAM[C]//2022 IEEE 6th Information Technology and Mechatronics Engineering Conference (ITOEC). Piscataway: IEEE, 2022: 386-391.
26	Rublee E, Rabaud V, Konolige K, et al. ORB: An Efficient Alternative to SIFT or SURF[C]//2011 International Conference on Computer Vision. Piscataway: IEEE, 2011: 2564-2571.
27	You Yingxuan, Wei Peng, Cai Jialun, et al. MISD-SLAM: Multimodal Semantic SLAM for Dynamic Environments[J]. Wireless Communications and Mobile Computing, 2022, 2022(1): 7600669.
28	Andréa Macario Barros, Michel Maugan, Moline Yoann, et al. A Comprehensive Survey of Visual SLAM Algorithms[J]. Robotics, 2022, 11(1): 24.
29	Chi Ming Chung, Yang Che Tseng, Ching Hsu Ya, et al. Orbeez-SLAM: A Real-time Monocular Visual SLAM with ORB Features and NeRF-realized Mapping[C]//2023 IEEE International Conference on Robotics and Automation (ICRA). Piscataway: IEEE, 2023: 9400-9406.
30	Campos Carlos, Elvira Richard, J Gómez Rodríguez Juan, et al. ORB-SLAM3: An Accurate Open-source Library for Visual, Visual-inertial, and Multimap SLAM[J]. IEEE Transactions on Robotics, 2021, 37(6): 1874-1890.
31	Neira José, Tardós Juan D. Data Association in Stochastic Mapping Using the Joint Compatibility Test[J]. IEEE Transactions on Robotics and Automation, 2001, 17(6): 890-897.
32	Rao Dan, Yuan Yuhao, Chen Xin. Research on Feature Extraction and Matching Algorithm in Visual SLAM[C]//2021 IEEE 5th Advanced Information Technology, Electronic and Automation Control Conference (IAEAC). Piscataway: IEEE, 2021: 522-526.
33	Watson A B. Digital Images and Human Vision[M]. Cambridge: MIT Press, 1993.
34	Ma Chaoqun, Hu Xiaoguang, Xiao Jin, et al. Homogenized ORB Algorithm Using Dynamic Threshold and Improved Quadtree[J]. Mathematical Problems in Engineering, 2021, 2021(1): 6693627.
35	Leutenegger Stefan, Chli Margarita, Siegwart Roland Y. BRISK: Binary Robust Invariant Scalable Keypoints[C]//2011 International Conference on Computer Vision. Piscataway: IEEE, 2011: 2548-2555.
36	Alcantarilla P F, Solutions T. Fast Explicit Diffusion for Accelerated Features in Nonlinear Scale Spaces[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 34(7): 1281-1298.
37	Burri Michael, Nikolic Janosch, Gohl Pascal, et al. The EuRoC Micro Aerial Vehicle Datasets[J]. The International Journal of Robotics Research, 2016, 35(10): 1157-1163.
38	David Zuñiga Noël, Jaenal Alberto, Ruben Gomez Ojeda, et al. The UMA-VI Dataset: Visual-inertial Odometry in Low-textured and Dynamic Illumination Environments[J]. The International Journal of Robotics Research, 2020, 39(9): 1052-1060.
39	Sturm Jürgen, Engelhard Nikolas, Endres Felix, et al. A Benchmark for the Evaluation of RGB-D SLAM Systems[C]//2012 IEEE/RSJ International Conference on Intelligent Robots and Systems. Piscataway: IEEE, 2012: 573-580.
40	Ferrera Maxime, Eudes Alexandre, Moras Julien, et al. OV² SLAM: A Fully Online and Versatile Visual SLAM for Real-time Applications[J]. IEEE Robotics and Automation Letters, 2021, 6(2): 1399-1406.
41	Taranco Raúl, Arnau José-Maria, González Antonio. LOCATOR: Low-power ORB Accelerator for Autonomous Cars[J]. Journal of Parallel and Distributed Computing, 2023, 174: 32-45.

[1]	邵绪强, 程雅, 金佚钟. 基于聚类融合的三维流线可视化方法[J]. 系统仿真学报, 2024, 36(3): 625-635.
[2]	符强, 腾先云, 纪元法, 任风华. 基于改进特征描述的SLAM动态算法研究[J]. 系统仿真学报, 2024, 36(11): 2712-2721.
[3]	罗宇成, 张明恩, 刘飞, 陆营波, 叶丰. 基于分段特征提取的仿真模型结果验证方法[J]. 系统仿真学报, 2024, 36(1): 272-281.
[4]	张凤全, 曹铎, 马晓寒, 陈柏君, 张江霄. 一种面向戏曲妆容细节生成的风格迁移网络[J]. 系统仿真学报, 2023, 35(9): 2064-2076.
[5]	刘杨杨, 丁刚毅, 闫大鹏, 薛彤. 超高清视频纹理的实时仿真方法[J]. 系统仿真学报, 2023, 35(8): 1748-1756.
[6]	罗慧诚, 汪淑娟. 基于特征鲁棒性增强的多摄像头下车辆识别方法[J]. 系统仿真学报, 2023, 35(5): 1059-1074.
[7]	金炜东, 张述礼, 唐鹏, 张曼. 基于稠密残差块与通道像素注意力的图像去雾网络[J]. 系统仿真学报, 2022, 34(8): 1663-1673.
[8]	祝志超, 吴定会, 岳远昌. 基于监督核熵成分分析的发动机磨损故障诊断[J]. 系统仿真学报, 2022, 34(1): 45-52.
[9]	于雅楠, 史敦煌, 华春杰. 特征点法SLAM视觉里程计自适应优化算法[J]. 系统仿真学报, 2022, 34(1): 104-112.
[10]	车辚辚, 孔英会, 陈智雄. 基于最大方差展开法的电能质量扰动识别[J]. 系统仿真学报, 2019, 31(8): 1702-1710.
[11]	王天润, 苏中, 刘宁, 李超, 付国栋. 跌倒检测的时频特征提取方法研究[J]. 系统仿真学报, 2019, 31(12): 2600-2605.
[12]	贺梦嘉, 吴迎年, 杨睿. 基于改进深度回归网络的无创血糖检测算法研究[J]. 系统仿真学报, 2019, 31(11): 2492-2498.
[13]	李晋芳, 李日福, 何汉武, 许军龙, 刘家远, 李航. 脸部整形手术仿真中的网格变形处理[J]. 系统仿真学报, 2018, 30(7): 2453-2458.
[14]	周颐, 孟明, 吴威, 周忠. 基于视频模型的虚拟现实视频融合系统[J]. 系统仿真学报, 2018, 30(7): 2550-2557.
[15]	李乾, 高尚兵, 潘志庚, 张正伟, 方澄华, 王圣全. 基于无标记识别的增强现实方法研究[J]. 系统仿真学报, 2018, 30(7): 2608-2614.