來源:83 項開源視覺 SLAM 方案夠你用了嗎?
前言
1. 本文由知乎作者小吳同學同步發佈於https://zhuanlan.zhihu.com/p/115599978/並持續更新。
2. 本文簡單將各種開源視覺SLAM方案分為以下 7 類(固然有不少文章無法恰當分類):
·Geometric SLAM
·Semantic / Learning SLAM
·Multi-Landmarks / Object SLAM
·VIO / VISLAM
·Dynamic SLAM
·Mapping
·Optimization
3. 由於本人自 2019 年 3 月開始整理,所以以下的代碼除了經典的框架之外基本都集中在 19-20 年;此外個人比較關注 VO、物體級 SLAM 和多路標 SLAM,所以以下內容收集的也不完整,無法涵蓋視覺 SLAM 的所有研究,僅作參考。
一、Geometric SLAM(20 項)
這一類是傳統的基於特徵點、直接法或半直接法的 SLAM,雖說傳統,但 2019 年也新誕生了 9 個開源方案。
1. PTAM
論文:Klein G, Murray D. Parallel tracking and mapping for small AR workspaces[C]//Mixed andAugmented Reality, 2007. ISMAR 2007. 6th IEEE and ACM International Symposiumon. IEEE, 2007: 225-234.
代碼:https://github.com/Oxford-PTAM/PTAM-GPL
工程地址:http://www.robots.ox.ac.uk/~gk/PTAM/
作者其他研究:http://www.robots.ox.ac.uk/~gk/publications.html
2. S-PTAM(雙目 PTAM)
論文:Taihú Pire,Thomas Fischer, Gastón Castro,Pablo De Cristóforis, Javier Civera and Julio Jacobo Berlles. S-PTAM: Stereo Parallel Tracking and Mapping. Robotics and AutonomousSystems, 2017.
代碼:https://github.com/lrse/sptam
作者其他論文:Castro G,Nitsche M A, Pire T, et al. Efficient on-board Stereo SLAM throughconstrained-covisibility strategies[J]. Robotics and Autonomous Systems, 2019.
3. MonoSLAM
論文:Davison A J, Reid I D, Molton N D, et al. MonoSLAM:Real-time single camera SLAM[J]. IEEE transactions on patternanalysis and machine intelligence, 2007, 29(6): 1052-1067.
代碼:https://github.com/hanmekim/SceneLib2
4. ORB-SLAM2
論文:Mur-Artal R, Tardós J D. Orb-slam2: Anopen-source slam system for monocular, stereo, and rgb-d cameras[J]. IEEETransactions on Robotics, 2017, 33(5): 1255-1262.
代碼:https://github.com/raulmur/ORB_SLAM2
單目半稠密建圖:Mur-Artal R, Tardós J D. Probabilistic Semi-Dense Mapping from Highly AccurateFeature-Based Monocular SLAM[C]//Robotics: Science and Systems. 2015,2015.
VIORB:Mur-Artal R, Tardós J D. Visual-inertialmonocular SLAM with map reuse[J]. IEEE Robotics and AutomationLetters, 2017, 2(2): 796-803.
多地圖:Elvira R, Tardós J D, Montiel J M M. ORBSLAM-Atlas: arobust and accurate multi-map system[J]. arXiv preprint arXiv:1908.11585, 2019.
以下 5, 6, 7, 8 幾項是 TUM 計算機視覺組全家桶
5. DSO
論文:Engel J, Koltun V, Cremers D. Direct sparseodometry[J]. IEEE transactions on pattern analysis and machineintelligence, 2017, 40(3): 611-625.
代碼:https://github.com/JakobEngel/dso
雙目 DSO:Wang R, Schworer M, Cremers D. Stereo DSO: Large-scale direct sparse visual odometry withstereo cameras[C]//Proceedings of the IEEE International Conference onComputer Vision. 2017: 3903-3911.
VI-DSO:Von Stumberg L, Usenko V, Cremers D. Direct sparsevisual-inertial odometry using dynamic marginalization[C]//2018 IEEEInternational Conference on Robotics and Automation (ICRA). IEEE, 2018:2510-2517.
6. LDSO
高翔在 DSO 上添加閉環的工作
論文:Gao X, Wang R, Demmel N, et al. LDSO: Directsparse odometry with loop closure[C]//2018 IEEE/RSJ InternationalConference on Intelligent Robots and Systems (IROS). IEEE, 2018:2198-2204.
代碼:https://github.com/tum-vision/LDSO
7. LSD-SLAM
論文:Engel J, Schöps T, Cremers D. LSD-SLAM: Large-scale direct monocular SLAM[C]//Europeanconference on computer vision. Springer, Cham, 2014: 834-849.
代碼:https://github.com/tum-vision/lsd_slam
8. DVO-SLAM
論文:Kerl C, Sturm J, Cremers D. Dense visualSLAM for RGB-D cameras[C]//2013 IEEE/RSJ International Conferenceon Intelligent Robots and Systems. IEEE, 2013: 2100-2106.
代碼 1:https://github.com/tum-vision/dvo_slam
代碼 2:https://github.com/tum-vision/dvo
Kerl C, Sturm J,Cremers D. Robust odometry estimation for RGB-D cameras[C]//2013 IEEEinternational conference on robotics and automation. IEEE, 2013:3748-3754.
Steinbrücker F,Sturm J, Cremers D. Real-time visual odometry from dense RGB-D images[C]//2011 IEEEinternational conference on computer vision workshops (ICCV Workshops). IEEE, 2011:719-722.
9. SVO
蘇黎世大學機器人與感知課題組
論文:Forster C, Pizzoli M, Scaramuzza D. SVO: Fast semi-direct monocular visual odometry[C]//2014 IEEEinternational conference on robotics and automation (ICRA). IEEE, 2014:15-22.
代碼:https://github.com/uzh-rpg/rpg_svo
Forster C, ZhangZ, Gassner M, et al. SVO: Semidirect visual odometry for monocular andmulticamera systems[J]. IEEE Transactions on Robotics, 2016,33(2): 249-265.
10. DSM
論文:Zubizarreta J, Aguinaga I, Montiel J M M. Direct sparsemapping[J]. arXiv preprint arXiv:1904.06577, 2019.
代碼:https://github.com/jzubizarreta/dsm
11. openvslam
論文:Sumikura S,Shibuya M, Sakurada K. OpenVSLAM: A Versatile Visual SLAM Framework[C]//Proceedingsof the 27th ACM International Conference on Multimedia. 2019: 2292-2295.
代碼:https://github.com/xdspacelab/openvslam
12. se2lam(地面車輛位姿估計的視覺里程計)
論文:Zheng F, Liu Y H. Visual-OdometricLocalization and Mapping for Ground Vehicles Using SE (2)-XYZ Constraints[C]//2019International Conference on Robotics and Automation (ICRA). IEEE, 2019:3556-3562.
代碼:https://github.com/izhengfan/se2lam
作者的另外一項工作
論文:Zheng F, Tang H,Liu Y H. Odometry-vision-basedground vehicle motion estimation with se (2)-constrained se (3) poses[J]. IEEEtransactions on cybernetics, 2018, 49(7): 2652-2663.
代碼:https://github.com/izhengfan/se2clam
13. GraphSfM(基於圖的並行大尺度 SFM)
論文:Chen Y, Shen S,Chen Y, et al. Graph-BasedParallel Large Scale Structure from Motion[J]. arXivpreprint arXiv:1912.10659, 2019.
代碼:https://github.com/AIBluefisher/GraphSfM
14. LCSD_SLAM(松耦合的半直接法單目 SLAM)
論文:Lee S H, Civera J. Loosely-Coupledsemi-direct monocular SLAM[J]. IEEE Robotics and AutomationLetters, 2018, 4(2): 399-406.
代碼:https://github.com/sunghoon031/LCSD_SLAM;谷歌學術 ;演示視頻
作者另外一篇關於單目尺度的文章代碼開源:Lee S H, deCroon G. Stability-based scale estimation for monocular SLAM[J]. IEEERobotics and Automation Letters, 2018, 3(2): 780-787.
15. RESLAM(基於邊的 SLAM)
論文:Schenk F, Fraundorfer F. RESLAM: Areal-time robust edge-based SLAM system[C]//2019 International Conference onRobotics and Automation (ICRA). IEEE, 2019: 154-160.
代碼:https://github.com/fabianschenk/RESLAM
16. scale_optimization(將單目 DSO 拓展到雙目)
論文:Mo J, Sattar J. ExtendingMonocular Visual Odometry to Stereo Camera System by Scale Optimization[C].International Conference on Intelligent Robots and Systems (IROS), 2019.
代碼:https://github.com/jiawei-mo/scale_optimization
17. BAD-SLAM(直接法 RGB-D SLAM)
論文:Schops T, Sattler T, Pollefeys M. BAD SLAM: Bundle Adjusted Direct RGB-D SLAM[C]//Proceedingsof the IEEE Conference on Computer Vision and Pattern Recognition. 2019:134-144.
代碼:https://github.com/ETH3D/badslam
18. GSLAM(集成 ORB-SLAM2,DSO,SVO 的通用框架)
論文:Zhao Y, Xu S, Bu S, et al. GSLAM: A general SLAM framework and benchmark[C]//Proceedingsof the IEEE International Conference on Computer Vision. 2019:1110-1120.
代碼:https://github.com/zdzhaoyong/GSLAM
19. ARM-VO(運行於 ARM 處理器上的單目 VO)
論文:Nejad Z Z, Ahmadabadian A H. ARM-VO: an efficient monocular visual odometry for groundvehicles on ARM CPUs[J]. Machine Vision and Applications, 2019:1-10.
代碼:https://github.com/zanazakaryaie/ARM-VO
20. cvo-rgbd(直接法 RGB-D VO)
論文:Ghaffari M, Clark W, Bloch A, et al. ContinuousDirect Sparse Visual Odometry from RGB-D Images[J]. arXivpreprint arXiv:1904.02266, 2019.
代碼:https://github.com/MaaniGhaffari/cvo-rgbd
二、Semantic / Learning SLAM(12 項)
SLAM 與深度學習相結合的工作當前主要體現在兩個方面,一方面是將語義信息參與到建圖、位姿估計等環節中,另一方面是端到端地完成 SLAM 的某一個步驟(比如 VO,閉環等)。個人對後者沒太關注,也同樣歡迎大家在issue中分享。
21. MsakFusion
論文:Runz M, Buffier M, Agapito L. Maskfusion:Real-time recognition, tracking and reconstruction of multiple moving objects[C]//2018 IEEEInternational Symposium on Mixed and Augmented Reality (ISMAR). IEEE, 2018:10-20.
代碼:https://github.com/martinruenz/maskfusion
22. SemanticFusion
論文:McCormac J, Handa A, Davison A, et al. Semanticfusion:Dense 3d semantic mapping with convolutional neural networks[C]//2017 IEEEInternational Conference on Robotics and automation (ICRA). IEEE, 2017:4628-4635.
代碼:https://github.com/seaun163/semanticfusion
23. semantic_3d_mapping
論文:Yang S, Huang Y, Scherer S. Semantic 3Doccupancy mapping through efficient high order CRFs[C]//2017IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).IEEE, 2017: 590-597.
代碼:https://github.com/shichaoy/semantic_3d_mapping
24. Kimera(實時度量與語義定位建圖開源庫)
論文:Rosinol A, AbateM, Chang Y, et al. Kimera: anOpen-Source Library for Real-Time Metric-Semantic Localization and Mapping[J]. arXivpreprint arXiv:1910.02490, 2019.
代碼:https://github.com/MIT-SPARK/Kimera
25. NeuroSLAM(腦啟發式 SLAM)
論文:Yu F, Shang J, Hu Y, et al. NeuroSLAM: a brain-inspired SLAM system for 3Denvironments[J]. Biological Cybernetics, 2019: 1-31.
代碼:https://github.com/cognav/NeuroSLAM
第四作者就是 Rat SLAM 的作者,文章也比較了十餘種腦啟發式的 SLAM
26. gradSLAM(自動分區的稠密 SLAM)
論文:Jatavallabhula K M, Iyer G, Paull L. gradSLAM:Dense SLAM meets Automatic Differentiation[J]. arXivpreprint arXiv:1910.10672, 2019.
代碼(預計 20 年 4 月放出):https://github.com/montrealrobotics/gradSLAM
27. ORB-SLAM2 + 目標檢測/分割的方案語義建圖
https://github.com/floatlazer/semantic_slam
https://github.com/qixuxiang/orb-slam2_with_semantic_labelling
https://github.com/Ewenwan/ORB_SLAM2_SSD_Semantic
28. SIVO(語義輔助特徵選擇)
論文:Ganti P, Waslander S. NetworkUncertainty Informed Semantic Feature Selection for Visual SLAM[C]//2019 16thConference on Computer and Robot Vision (CRV). IEEE, 2019: 121-128.
代碼:https://github.com/navganti/SIVO
29. FILD(臨近圖增量式閉環檢測)
論文:Shan An, Guangfu Che, Fangru Zhou,Xianglong Liu, Xin Ma, Yu Chen.Fast and Incremental Loop Closure Detection usingProximity Graphs. pp. 378-385, The 2019 IEEE/RSJ International Conferenceon Intelligent Robots and Systems (IROS2019)
代碼:https://github.com/AnshanTJU/FILD
30. object-detection-sptam(目標檢測與雙目 SLAM)
論文:Pire T, Corti J, Grinblat G. Online Object Detection and Localization on Stereo VisualSLAM System[J]. Journal of Intelligent & Robotic Systems, 2019:1-10.
代碼:https://github.com/CIFASIS/object-detection-sptam
31. Map Slammer(單目深度估計 + SLAM)
論文:Torres-Camara J M, Escalona F, Gomez-DonosoF, et al. Map Slammer: Densifying Scattered KSLAM 3D Maps withEstimated Depth[C]//Iberian Robotics conference. Springer, Cham, 2019:563-574.
代碼:https://github.com/jmtc7/mapSlammer
32. NOLBO(變分模型的概率 SLAM)
論文:Yu H, Lee B. Not Only LookBut Observe: Variational Observation Model of Scene-Level 3D Multi-ObjectUnderstanding for Probabilistic SLAM[J]. arXiv preprint arXiv:1907.09760, 2019.
代碼:https://github.com/bogus2000/NOLBO
三、Multi-Landmarks / Object SLAM(12 項)
其實多路標的點、線、平面 SLAM 和物體級 SLAM 完全可以分類在 Geometric SLAM 和 Semantic SLAM中,但個人對這一方向比較感興趣(也是我的研究生課題),所以將其獨立出來,開源方案相對較少,但很有意思。
33. PL-SVO(點線 SVO)
論文:Gomez-Ojeda R, Briales J, Gonzalez-JimenezJ. PL-SVO: Semi-direct Monocular Visual Odometry by combiningpoints and line segments[C]//Intelligent Robots and Systems(IROS), 2016 IEEE/RSJ International Conference on. IEEE, 2016:4211-4216.
代碼:https://github.com/rubengooj/pl-svo
34. stvo-pl(雙目點線 VO)
論文:Gomez-Ojeda R, Gonzalez-Jimenez J. Robust stereo visual odometry through a probabilisticcombination of points and line segments[C]//2016 IEEE International Conferenceon Robotics and Automation (ICRA). IEEE, 2016: 2521-2526.
代碼:https://github.com/rubengooj/stvo-pl
35. PL-SLAM(點線 SLAM)
論文:Gomez-Ojeda R, Zuñiga-Noël D, Moreno F A,et al. PL-SLAM: aStereo SLAM System through the Combination of Points and Line Segments[J]. arXivpreprint arXiv:1705.09479, 2017.
代碼:https://github.com/rubengooj/pl-slam
Gomez-Ojeda R,Moreno F A, Zuñiga-Noël D, et al.PL-SLAM: a stereo SLAM system through the combination ofpoints and line segments[J]. IEEE Transactions on Robotics, 2019,35(3): 734-746.
36. PL-VIO
論文:He Y, Zhao J, Guo Y, et al. PL-VIO:Tightly-coupled monocular visual–inertial odometry using point and linefeatures[J]. Sensors, 2018, 18(4): 1159.
代碼:https://github.com/HeYijia/PL-VIO
VINS + 線段:https://github.com/Jichao-Peng/VINS-Mono-Optimization
37. lld-slam(用於 SLAM 的可學習型線段描述符)
論文:Vakhitov A, Lempitsky V. Learnable line segment descriptor for visual SLAM[J]. IEEEAccess, 2019, 7: 39923-39934.
代碼:https://github.com/alexandervakhitov/lld-slam;Video
點線結合的工作還有很多,國內的比如 + 上交鄒丹平老師的 Zou D, Wu Y, Pei L, et al. StructVIO:visual-inertial odometry with structural regularity of man-made environments[J]. IEEETransactions on Robotics, 2019, 35(4): 999-1013. + 浙大的 Zuo X, Xie X, Liu Y, et al. Robust visualSLAM with point and line features[C]//2017 IEEE/RSJ InternationalConference on Intelligent Robots and Systems (IROS). IEEE, 2017:1775-1782.
38. PlaneSLAM
論文:Wietrzykowski J. On the representation of planes for efficient graph-basedslam with high-level features[J]. Journal of Automation MobileRobotics and Intelligent Systems, 2016, 10.
代碼:https://github.com/LRMPUT/PlaneSLAM
作者另外一項開源代碼,沒有找到對應的論文:https://github.com/LRMPUT/PUTSLAM
39. Eigen-Factors(特徵因子平面對齊)
論文:Ferrer G. Eigen-Factors: Plane Estimation for Multi-Frame andTime-Continuous Point Cloud Alignment[C]//2019 IEEE/RSJ InternationalConference on Intelligent Robots and Systems (IROS). IEEE, 2019:1278-1284.
代碼:https://gitlab.com/gferrer/eigen-factors-iros2019
40. PlaneLoc
論文:Wietrzykowski J, Skrzypczyński P. PlaneLoc:Probabilistic global localization in 3-D using local planar features[J]. Roboticsand Autonomous Systems, 2019, 113: 160-173.
代碼:https://github.com/LRMPUT/PlaneLoc
41. Pop-up SLAM
論文:Yang S, Song Y, Kaess M, et al. Pop-up slam:Semantic monocular plane slam for low-texture environments[C]//2016IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).IEEE, 2016: 1222-1229.
代碼:https://github.com/shichaoy/pop_up_slam
42. Object SLAM
論文:Mu B, Liu S Y, Paull L, et al. Slam withobjects using a nonparametric pose graph[C]//2016 IEEE/RSJ InternationalConference on Intelligent Robots and Systems (IROS). IEEE, 2016:4602-4609.
代碼:https://github.com/BeipengMu/objectSLAM
43. voxblox-plusplus(物體級體素建圖)
論文:Grinvald M, Furrer F, Novkovic T, et al. Volumetricinstance-aware semantic mapping and 3D object discovery[J]. IEEERobotics and Automation Letters, 2019, 4(3): 3037-3044.
代碼:https://github.com/ethz-asl/voxblox-plusplus
44. Cube SLAM
論文:Yang S, Scherer S. Cubeslam:Monocular 3-d object slam[J]. IEEE Transactions on Robotics, 2019,35(4): 925-938.
代碼:https://github.com/shichaoy/cube_slam
也有很多有意思的但沒開源的物體級 SLAM
Ok K, Liu K,Frey K, et al. RobustObject-based SLAM for High-speed Autonomous Navigation[C]//2019International Conference on Robotics and Automation (ICRA). IEEE, 2019:669-675.
Li J, Meger D,Dudek G. SemanticMapping for View-Invariant Relocalization[C]//2019International Conference on Robotics and Automation (ICRA). IEEE, 2019:7108-7115.
Nicholson L,Milford M, Sünderhauf N. Quadricslam:Dual quadrics from object detections as landmarks in object-oriented slam[J]. IEEERobotics and Automation Letters, 2018, 4(1): 1-8.
四、VIO / VISLAM(10 項)
在傳感器融合方面只關注了視覺 + 慣導,其他傳感器像 LiDAR,GPS 關注較少(SLAM 太複雜啦 -_-! )。視慣融合的新工作也相對較少,基本一些經典的方案就夠用了。
45. msckf_vio
論文:Sun K, Mohta K, Pfrommer B, et al. Robust stereovisual inertial odometry for fast autonomous flight[J]. IEEERobotics and Automation Letters, 2018, 3(2): 965-972.
代碼:https://github.com/KumarRobotics/msckf_vio
46. rovio
論文:Bloesch M, Omari S, Hutter M, et al. Robust visual inertial odometry using a direct EKF-basedapproach[C]//2015 IEEE/RSJ international conference onintelligent robots and systems (IROS). IEEE, 2015: 298-304.
代碼:https://github.com/ethz-asl/rovio
47. R-VIO
論文:Huai Z, Huang G. Robocentricvisual-inertial odometry[C]//2018 IEEE/RSJ InternationalConference on Intelligent Robots and Systems (IROS). IEEE, 2018:6319-6326.
代碼:https://github.com/rpng/R-VIO
48. okvis
論文:Leutenegger S, Lynen S, Bosse M, et al. Keyframe-based visual–inertial odometry using nonlinearoptimization[J]. The International Journal of Robotics Research, 2015,34(3): 314-334.
代碼:https://github.com/ethz-asl/okvis
49. VIORB
論文:Mur-Artal R, Tardós J D. Visual-inertialmonocular SLAM with map reuse[J]. IEEE Robotics and AutomationLetters, 2017, 2(2): 796-803.
代碼:https://github.com/jingpang/LearnVIORB(VIORB 本身是沒有開源的,這是王京大佬復現的一個版本)
50. VINS-mono
論文:Qin T, Li P, Shen S. Vins-mono: Arobust and versatile monocular visual-inertial state estimator[J]. IEEETransactions on Robotics, 2018, 34(4): 1004-1020.
代碼:https://github.com/HKUST-Aerial-Robotics/VINS-Mono
雙目版 VINS-Fusion:https://github.com/HKUST-Aerial-Robotics/VINS-Fusion
移動段 VINS-mobile:https://github.com/HKUST-Aerial-Robotics/VINS-Mobile
51. VINS-RGBD
論文:Shan Z, Li R, Schwertfeger S. RGBD-InertialTrajectory Estimation and Mapping for Ground Robots[J]. Sensors, 2019,19(10): 2251.
代碼:https://github.com/STAR-Center/VINS-RGBD
52. Open-VINS
論文:Geneva P, Eckenhoff K, Lee W, et al. Openvins: A research platform for visual-inertialestimation[C]//IROS 2019 Workshop on Visual-Inertial Navigation:Challenges and Applications, Macau, China. IROS 2019.
代碼:https://github.com/rpng/open_vins
53. versavis(多功能的視慣傳感器系統)
論文:Tschopp F, RinerM, Fehr M, et al. VersaVIS—AnOpen Versatile Multi-Camera Visual-Inertial Sensor Suite[J]. Sensors, 2020,20(5): 1439.
代碼:https://github.com/ethz-asl/versavis
54. CPI(視慣融合的封閉式預積分)
論文:Eckenhoff K, Geneva P, Huang G. Closed-form preintegration methods for graph-basedvisual–inertial navigation[J]. The International Journal ofRobotics Research, 2018.
代碼:https://github.com/rpng/cpi
五、Dynamic SLAM(5 項)
動態 SLAM 也是一個很值得研究的話題,這裡不太好分類,很多工作用到了語義信息或者用來三維重建,收集的方案相對較少,歡迎補充issue。
55. DynamicSemanticMapping(動態語義建圖)
論文:Kochanov D, Ošep A, Stückler J, et al. Scene flow propagation for semantic mapping and objectdiscovery in dynamic street scenes[C]//Intelligent Robots and Systems(IROS), 2016 IEEE/RSJ International Conference on. IEEE, 2016:1785-1792.
代碼:https://github.com/ganlumomo/DynamicSemanticMapping
56. DS-SLAM(動態語義 SLAM)
論文:Yu C, Liu Z, Liu X J, et al. DS-SLAM: Asemantic visual SLAM towards dynamic environments[C]//2018IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).IEEE, 2018: 1168-1174.
代碼:https://github.com/ivipsourcecode/DS-SLAM
57. Co-Fusion(實時分割與跟蹤多物體)
論文:Rünz M, Agapito L. Co-fusion:Real-time segmentation, tracking and fusion of multiple objects[C]//2017 IEEEInternational Conference on Robotics and Automation (ICRA). IEEE, 2017:4471-4478.
代碼:https://github.com/martinruenz/co-fusion
58. DynamicFusion
論文:Newcombe R A, Fox D, Seitz S M. Dynamicfusion: Reconstruction and tracking of non-rigidscenes in real-time[C]//Proceedings of the IEEE conference oncomputer vision and pattern recognition. 2015: 343-352.
代碼:https://github.com/mihaibujanca/dynamicfusion
59. ReFusion(動態場景利用殘差三維重建)
論文:Palazzolo E, Behley J, Lottes P, et al. ReFusion: 3DReconstruction in Dynamic Environments for RGB-D Cameras Exploiting Residuals[J]. arXivpreprint arXiv:1905.02082, 2019.
代碼:https://github.com/PRBonn/refusion
六、Mapping(18 項)
針對建圖的工作一方面是利用幾何信息進行稠密重建,另一方面很多工作利用語義信息達到了很好的語義重建效果,三維重建本身就是個很大的話題,開源代碼也很多,以下方案收集地可能也不太全。
60. InfiniTAM(跨平臺 CPU
實時重建)論文:Prisacariu V A,Kähler O, Golodetz S, et al. Infinitam v3: A framework for large-scale 3dreconstruction with loop closure[J]. arXiv preprint arXiv:1708.00783, 2017.
代碼:https://github.com/victorprad/InfiniTAM
61. BundleFusion
論文:Dai A, Nießner M, Zollhöfer M, et al. Bundlefusion:Real-time globally consistent 3d reconstruction using on-the-fly surfacereintegration[J]. ACM Transactions on Graphics (TOG), 2017,36(4): 76a.
代碼:https://github.com/niessner/BundleFusion
62. KinectFusion
論文:Newcombe R A, Izadi S, Hilliges O, et al. KinectFusion: Real-time dense surface mapping and tracking[C]//2011 10thIEEE International Symposium on Mixed and Augmented Reality. IEEE, 2011:127-136.
代碼:https://github.com/chrdiller/KinectFusionApp
63. ElasticFusion
論文:Whelan T, Salas-Moreno R F, Glocker B, etal. ElasticFusion: Real-time dense SLAM and light sourceestimation[J]. The International Journal of Robotics Research, 2016,35(14): 1697-1716.
代碼:https://github.com/mp3guy/ElasticFusion
64. Kintinuous
ElasticFusion 同一個團隊的工作,帝國理工 Stefan Leutenegger
論文:Whelan T, Kaess M, Johannsson H, et al. Real-time large-scale dense RGB-D SLAM with volumetricfusion[J]. The International Journal of Robotics Research, 2015,34(4-5): 598-626.
代碼:https://github.com/mp3guy/Kintinuous
65. ElasticReconstruction
論文:Choi S, Zhou Q Y, Koltun V. Robust reconstruction of indoor scenes[C]//Proceedingsof the IEEE Conference on Computer Vision and Pattern Recognition. 2015:5556-5565.
代碼:https://github.com/qianyizh/ElasticReconstruction
66. FlashFusion
論文:Han L, Fang L. FlashFusion:Real-time Globally Consistent Dense 3D Reconstruction using CPU Computing[C]. RSS, 2018.
代碼(一直沒放出來):https://github.com/lhanaf/FlashFusion
67. RTAB-Map(激光視覺稠密重建)
論文:Labbé M, Michaud F. RTAB‐Map as an open‐source lidar and visual simultaneouslocalization and mapping library for large‐scale and long‐term online operation[J]. Journal ofField Robotics, 2019, 36(2): 416-446.
代碼:https://github.com/introlab/rtabmap
68. RobustPCLReconstruction(戶外稠密重建)
論文:Lan Z, Yew Z J, Lee G H. Robust Point Cloud Based Reconstruction of Large-ScaleOutdoor Scenes[C]//Proceedings of the IEEE Conference on ComputerVision and Pattern Recognition. 2019: 9690-9698.
代碼:https://github.com/ziquan111/RobustPCLReconstruction
69. plane-opt-rgbd(室內平面重建)
論文:Wang C, Guo X. Efficient Plane-Based Optimization of Geometry and Texturefor Indoor RGB-D Reconstruction[C]//Proceedings of the IEEE Conferenceon Computer Vision and Pattern Recognition Workshops. 2019: 49-53.
代碼:https://github.com/chaowang15/plane-opt-rgbd
70. DenseSurfelMapping(稠密表面重建)
論文:Wang K, Gao F, Shen S. Real-timescalable dense surfel mapping[C]//2019 International Conference onRobotics and Automation (ICRA). IEEE, 2019: 6919-6925.
代碼:https://github.com/HKUST-Aerial-Robotics/DenseSurfelMapping
71. surfelmeshing(網格重建)
論文:Schöps T, Sattler T, Pollefeys M. Surfelmeshing:Online surfel-based mesh reconstruction[J]. IEEE Transactions on PatternAnalysis and Machine Intelligence, 2019.
代碼:https://github.com/puzzlepaint/surfelmeshing
72. DPPTAM(單目稠密重建)
論文:Concha Belenguer A, Civera Sancho J. DPPTAM: Dense piecewise planar tracking and mapping from amonocular sequence[C]//Proc. IEEE/RSJ Int. Conf. Intell. Rob. Syst. 2015(ART-2015-92153).
代碼:https://github.com/alejocb/dpptam
相關研究:基於超像素的單目 SLAM:UsingSuperpixels in Monocular SLAM ICRA 2014 ;谷歌學術
73. VI-MEAN(單目視慣稠密重建)
論文:Yang Z, Gao F, Shen S. Real-time monocular dense mapping on aerial robots usingvisual-inertial fusion[C]//2017 IEEE International Conference onRobotics and Automation (ICRA). IEEE, 2017: 4552-4559.
代碼:https://github.com/dvorak0/VI-MEAN
74. REMODE(單目概率稠密重建)
論文:Pizzoli M, Forster C, Scaramuzza D. REMODE: Probabilistic, monocular dense reconstruction inreal time[C]//2014 IEEE International Conference on Robotics andAutomation (ICRA). IEEE, 2014: 2609-2616.
原始開源代碼:https://github.com/uzh-rpg/rpg_open_remode
與 ORB-SLAM2 結合版本:https://github.com/ayushgaud/ORB_SLAM2https://github.com/ayushgaud/ORB_SLAM2
75. DeepFactors(實時的概率單目稠密 SLAM)
帝國理工學院戴森機器人實驗室
論文:Czarnowski J, Laidlow T, Clark R, et al. DeepFactors:Real-Time Probabilistic Dense Monocular SLAM[J]. arXivpreprint arXiv:2001.05049, 2020.
代碼:https://github.com/jczarnowski/DeepFactors(還未放出)
其他論文:Bloesch M,Czarnowski J, Clark R, et al. CodeSLAM—learning a compact, optimisable representationfor dense visual SLAM[C]//Proceedings of the IEEE conference oncomputer vision and pattern recognition. 2018: 2560-2568.
76. probabilistic_mapping(單目概率稠密重建)
港科沈邵劼老師團隊
論文:Ling Y, Wang K, Shen S. Probabilisticdense reconstruction from a moving camera[C]//2018IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).IEEE, 2018: 6364-6371.
代碼:https://github.com/ygling2008/probabilistic_mapping
另外一篇稠密重建文章的代碼一直沒放出來Github:Ling Y, Shen S. Real‐timedense mapping for online processing and navigation[J]. Journal ofField Robotics, 2019, 36(5): 1004-1036.
77. ORB-SLAM2 單目半稠密建圖
論文:Mur-Artal R, Tardós J D. Probabilistic Semi-Dense Mapping from Highly AccurateFeature-Based Monocular SLAM[C]//Robotics: Science and Systems. 2015,2015.
代碼(本身沒有開源,賀博復現的一個版本):https://github.com/HeYijia/ORB_SLAM2
加上線段之後的半稠密建圖
論文:He S, Qin X, Zhang Z, et al. Incremental3d line segment extraction from semi-dense slam[C]//2018 24thInternational Conference on Pattern Recognition (ICPR). IEEE, 2018:1658-1663.
代碼:https://github.com/shidahe/semidense-lines
作者在此基礎上用於指導遠程抓取操作的一項工作:https://github.com/atlas-jj/ORB-SLAM-free-space-carving
七、Optimization(6 項)
個人感覺優化可能是 SLAM 中最難的一部分了吧 +_+ ,我們一般都是直接用現成的因子圖、圖優化方案,要創新可不容易,分享山川小哥d的入坑指南https://zhuanlan.zhihu.com/p/53972892。
78. 後端優化庫
GTSAM:https://github.com/borglab/gtsam
g2o:https://github.com/RainerKuemmerle/g2o
ceres:http://ceres-solver.org/
79. ICE-BA
論文:Liu H, Chen M, Zhang G, et al. Ice-ba: Incremental, consistent and efficient bundleadjustment for visual-inertial slam[C]//Proceedings of the IEEE Conferenceon Computer Vision and Pattern Recognition. 2018: 1974-1982.
代碼:https://github.com/baidu/ICE-BA
80. minisam(因子圖最小二乘優化框架)
論文:Dong J, Lv Z. miniSAM: AFlexible Factor Graph Non-linear Least Squares Optimization Framework[J]. arXivpreprint arXiv:1909.00903, 2019.
代碼:https://github.com/dongjing3309/minisam
81. SA-SHAGO(幾何基元圖優化)
論文:Aloise I, Della Corte B, Nardi F, et al. Systematic Handling of Heterogeneous Geometric Primitivesin Graph-SLAM Optimization[J]. IEEE Robotics and AutomationLetters, 2019, 4(3): 2738-2745.
代碼:https://srrg.gitlab.io/sashago-website/index.html#
82. MH-iSAM2(SLAM 優化器)
論文:Hsiao M, Kaess M. MH-iSAM2:Multi-hypothesis iSAM using Bayes Tree and Hypo-tree[C]//2019International Conference on Robotics and Automation (ICRA). IEEE, 2019:1274-1280.
代碼:https://bitbucket.org/rpl_cmu/mh-isam2_lib/src/master/
83. MOLA(用於定位和建圖的模塊化優化框架)
論文:Blanco-Claraco J L. A ModularOptimization Framework for Localization and Mapping[J]. Proc. ofRobotics: Science and Systems (RSS), FreiburgimBreisgau, Germany, 2019,2.
代碼:https://github.com/MOLAorg/mola
閱讀更多 3D視覺工坊 的文章