The DOI matches its Crossref record; 62 of the 66 references it cites were found in a registry or an index.
Visual Odometry Revisited: What Should Be Learnt?
Open accessAbstract
In this work we present a monocular visual odometry (VO) algorithm which leverages geometry-based methods and deep learning. Most existing VO/SLAM systems with superior performance are based on geometry and have to be carefully designed for different application scenarios. Moreover, most monocular systems suffer from scale-drift issue. Some recent deep learning works learn VO in an end-to-end manner but the performance of these deep systems is still not comparable to geometry-based methods. In this work, we revisit the basics of VO and explore the right way for integrating deep learning with epipolar geometry and Perspective-n-Point (PnP) method. Specifically, we train two convolutional neural networks (CNNs) for estimating single-view depths and two-view optical flows as intermediate outputs. With the deep predictions, we design a simple but robust frame-to-frame VO algorithm (DF-VO) which outperforms pure deep learning-based and geometry-based methods. More importantly, our system does not suffer from the scale-drift issue being aided by a scale consistent single-view depth CNN. Extensive experiments on KITTI dataset shows the robustness of our system and a detailed ablation study shows the effect of different factors in our system. Code is available at here: DF-VO.
Overview
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“As shown in (Zhan et al. 2020), DF-VO with this correspondence selection scheme has already outperformed existing VO/SLAM baselines.”
From DF-VO: What Should Be Learnt for Visual Odometry? (2021) · cites this paper as Zhan et al. 2020
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
U-Net: Convolutional Networks for Biomedical Image Segmentation
Ronneberger, O., Fischer, P., & Brox, T. (2015) · Lecture notes in computer science · doi:10.1007/978-3-319-24574-4_28 · Cited by 99,404
Adam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946
Distinctive Image Features from Scale-Invariant Keypoints
Lowe, D. G. (2004) · International Journal of Computer Vision · doi:10.1023/b:visi.0000029664.99615.94 · Cited by 55,589
Multiple View Geometry in Computer Vision
Hartley, R., & Zisserman, A. (2004) · doi:10.1017/cbo9780511811685 · Cited by 21,070
Are we ready for autonomous driving? The KITTI vision benchmark suite
Geiger, A., Lenz, P., & Urtasun, R. (2012) · 2012 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2012.6248074 · Cited by 14,897
(2017) · Cited by 11,625
ORB: An efficient alternative to SIFT or SURF
Rublee, E., Rabaud, V., Konolige, K., & Bradski, G. (2011) · 2011 International Conference on Computer Vision · doi:10.1109/iccv.2011.6126544 · Cited by 11,037
Vision meets robotics: The KITTI dataset
Geiger, A., Lenz, P., Stiller, C., & Urtasun, R. (2013) · The International Journal of Robotics Research · doi:10.1177/0278364913491297 · Cited by 9,913
ORB-SLAM2: An Open-Source SLAM System for Monocular, Stereo, and RGB-D Cameras
Mur-Artal, R., & Tardos, J. D. (2017) · IEEE Transactions on Robotics · doi:10.1109/tro.2017.2705103 · Cited by 6,898 · cited 9×
“One of the most successful and accurate full SLAM systems using a sparse (ORB) feature-based approach is ORB-SLAM2 [13], along with DSO [16], a direct keyframe-based sparse SLAM method.”
MethodInfluentialORB-SLAM: A Versatile and Accurate Monocular SLAM System
Mur-Artal, R., Montiel, J. M. M., & Tardos, J. D. (2015) · IEEE Transactions on Robotics · doi:10.1109/tro.2015.2463671 · Cited by 6,778
Research works printed but not linked: 2
- godard, 2019, Digging into self-supervised monocular depth prediction, IEEE International Conference on Computer Vision (ICCV)
- 10.1023/A:1007941100561 · DOI
Cited by
Sorted by citations
Items 101–200 of 201
YOWO: You Only Walk Once to Jointly Map an Indoor Scene and Register Ceiling-Mounted Cameras
Yang, F., Yamao, S., Kusajima, I., Moteki, A., Masui, S., & Jiang, S. (2024) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2024.3453277 · Cited by 4
“In previous studies [13], [108], the Kabsch-Umeyama (KU) algorithm [109], [110] has been utilized to align relative CMC positions to the GT for evaluation purposes, or to realign relative CMC positions for iterative pose refinement.”
Depth-Consistent Monocular Visual Trajectory Estimation for AUVs
Wang, Y., Liu, X., Gu, D., Wang, J., & Fu, X. (2025) · IEEE Internet of Things Journal · doi:10.1109/jiot.2025.3527131 · Cited by 4
“Deep neural networks have powerful capabilities in mining featureless image information and predicting scene depth values [18], [19], [20].”
Kang, J. (2025) · Sensors · doi:10.3390/s25196050 · Cited by 4
“To address this limitation, a consistency-based filtering method is often employed [32].”
Zhou, S., Li, Z., Lv, Z., Zhou, C., Wu, P., Zhu, C., & Liu, W. (2023) · Sensors · doi:10.3390/s23239468 · Cited by 5
“This integration can also improve the inaccurate results of the visual algorithm caused by certain movements or scenes with simple textures [36].”
BackgroundMD2VO: Enhancing Monocular Visual Odometry through Minimum Depth Difference
Li, P., Tang, C., Duan, Y., & Li, Z. (2024) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn60899.2024.10649955 · Cited by 4
“Specifically, we compare MD2VO to two other methods: ORB-SLAM2 [4], a classic geometry-based monocular visual odometry method, and DF-VO [11], a learning-based method that also employs depth [31].”
BackgroundMethodInfluentialError Diagnosis of Deep Monocular Depth Estimation Models
Chawla, J., Thakurdesai, N., Godase, A., Reza, M. A., Crandall, D., & Jung, S. (2021) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros51168.2021.9636673 · Cited by 6
Camera pose estimation framework for array‐structured images
Shin, M.-J., Park, W., Kim, J. H., Kim, J., Yun, K., & Kang, S. (2022) · ETRI Journal · doi:10.4218/etrij.2021-0303 · Cited by 4
“The relative camera pose estimation between adjacent cameras is an essential low-level task for many computer vision applications, such as simultaneous localization and mapping (SLAM) [26 – 28], depth estimation [29 – 32], and visual odometry [33 – 35].”
BackgroundTDO-SLAM: Traffic Sign and Dynamic Object Based Visual SLAM
Park, S.-Y., & Lee, J. (2024) · IEEE Access · doi:10.1109/access.2024.3362675 · Cited by 3
“The evaluation tools [45], [46] are used for fair quantitative measurement.”
MethodVisual-Inertial Odometry Using Optical Flow from Deep Learning
Kang, J. M., Sjanic, Z., & Hendeby, G. (2024) · International Conference on Information Fusion (FUSION) · doi:10.23919/fusion59988.2024.10706322 · Cited by 3
“The idea pursued in [39] is that points for which the optical flow between the images I k and I k +1 and back, from I k +1 to I k , is consistent should have high quality flow.”
BackgroundAdaptive Visual-Aided 4D Radar Odometry Through Transformer-Based Feature Fusion
Zhang, Y., Xiao, R., Hong, Z., Hu, L., & Liu, J. (2024) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros58592.2024.10802367 · Cited by 3
“UndeepVO [12] and DF-VO [13] train depth and pose networks on calibrated stereo videos using a photometric loss.”
DualQuat-LOAM: LiDAR odometry and mapping parameterized on dual quaternions
Velasco-Sánchez, E., Recalde, L. F., Li, G., Candelas-Herías, F. A., Méndez, S. T. P., & Torres, F. (2025) · Robotics and Autonomous Systems · doi:10.1016/j.robot.2025.105009 · Cited by 3
“For the evaluation of our method, we employ the tools described in [50], which allow us to calculate the average translation (ATE) and rotation (ARE) errors for each sequence.”
Sensor Data Prediction in Missile Flight Tests
Ryu, S.-G., Jeong, J. J., & Shim, D. H. (2022) · Sensors · doi:10.3390/s22239410 · Cited by 3
“In visual odometry, the movement characteristics in images are effectively analyzed by optical flow, and the geometrical analysis combined with deep learning networks [42] improve performance.”
Tracking by segmentation with future motion estimation applied to person-following robots
Jiang, S., Cui, R., Wei, R., Fu, Z., Hong, Z., & Feng, G. (2023) · Frontiers in Neurorobotics · doi:10.3389/fnbot.2023.1255085 · Cited by 4
“approach enabled the proposed tracker to focus on a distinct target among similar foreground noises, effectively handling overlapping situations during tracking (Wang et al., 2019; Koide et al., 2020; Zhan et al., 2020).”
BackgroundDOR-SLAM: A Visual SLAM Based on Dynamic Object Removal for Dynamic Environments
Liao, G., & Yin, F. (2023) · China Automation Congress (CAC) · doi:10.1109/cac59555.2023.10451728 · Cited by 3
Sahoo, S. R., & Manivanan, P. V. (2024) · IEEE Access · doi:10.1109/access.2024.3424234 · Cited by 3
“The 2D LiDAR consists of a single scanning layer and performs well in indoor applications such as SLAM [9], [30].”
MethodSelf-supervised Pretraining and Finetuning for Monocular Depth and Visual Odometry
Antsfeld, L., & Chidlovskii, B. (2024) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra57147.2024.10611058 · Cited by 3
“Supervised learning methods have been applied in many deep neural network frameworks and demonstrated outstanding results in both visual odometry and depth estimation [6], [24], [47].”
MethodLyu, Y., Zhang, X., & Tao, F. (2024) · IEEE Transactions on Artificial Intelligence · doi:10.1109/tai.2024.3491950 · Cited by 3
“…[58], deep depth, deep pose, and deep uncertainty for monocular visual odometry (D3VO) [57], LiDAR-monocular visual odometry (LIMO) [15], Zhan [61], the approach name for joint direct stereo odometry (GDVO) [67], direct sparse visual odometry (DSO) [48], Huang [19], parameterized semantic…”
Deep Depth‐Flow Odometry With Inertial Sensor Fusion
Kang, J. (2025) · Electronics Letters · doi:10.1049/ell2.70409 · Cited by 3
Evaluating the Impact of Head Motion on Monocular Visual Odometry with Synthetic Data
Hamesse, C., Luong, H. Q., & Haelterman, R. (2022) · International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP) · doi:10.5220/0010881500003124 · Cited by 3
“Convolutional neural networks are used in the front-end for feature ex- traction, single-view depth (Yang et al., 2020) and/or two-view optical flow estimation (Zhan et al., 2020).”
MethodSensors and Data in Mobile Robotics for Localisation
Hodge, J. (2023) · IGI Global eBooks · doi:10.4018/978-1-7998-9220-5.ch133 · Cited by 3
Relative pose estimation from panoramic images using a hybrid neural network architecture
Offermann, L. (2024) · Scientific Reports · doi:10.1038/s41598-024-75124-7 · Cited by 2
“Notably, the works of Zhan et al. 36 and Yang et al. 37 show that combining traditional geometry-based methods for RPE with deep learning can lead to hybrid models that exceed the pose estimation quality of purely geometry-based and purely deep-learning based methods (see the work of Yang et al. 37…”
MethodDeepOF-VIO: A Filter-Based Visual-Inertial Odometry Using Deep Optical Flow
Kang, J. (2025) · IEEE Sensors Journal · doi:10.1109/jsen.2025.3639021 · Cited by 2
“A method for extracting high correspondences between bidirectional deep optical flows and estimating pose has been proposed [22].”
Review of deep learning-based visual SLAM: types, approaches, and future work
Zhao, L., Chen, T., Yuan, P., Li, X., & Chen, B. (2025) · Industrial Robot the international journal of robotics research and application · doi:10.1108/ir-04-2025-0137 · Cited by 2
CoProU-VO: Combining Projected Uncertainty for End-to-End Unsupervised Monocular Visual Odometry
Xie, J., Dhaouadi, O., Chen, W., Meier, J., Kaiser, J., & Cremers, D. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-12840-9_32 · Cited by 2
DPDM: Feature-Based Pose Refinement with Deep Pose and Deep Match for Monocular Visual Odometry
Huang, L.-Y., Huang, S.-S., & Chien, S. (2023) · Proceedings - International Conference on Image Processing · doi:10.1109/icip49359.2023.10221966 · Cited by 2
“Hybrid VO attempts to integrate deep learning and geometric constraints to gain advantages [3] [9].”
BackgroundMethodInfluentialYao, J., Li, Y., & Li, J. (2024) · Machine Vision and Applications · doi:10.1007/s00138-024-01586-4 · Cited by 2
“Several learning-based visual odometry methods aimed to directly recover the absolute scale in the scene [20–22].”
Global Localization using OpenStreetMap and Elevation Offsets
Przewodowski, A., Osório, F. S., & Grassi, V. (2024) · Journal of the Brazilian Computer Society · doi:10.5753/jbcs.2024.3795 · Cited by 2
OLM-SLAM: online lifelong memory system for simultaneous localization and mapping
Lu, H., Shen, Y., Zhang, Q., Jiang, Q., Zhu, Q., Fu, G., Xuemei, N., & Li, J. (2024) · Measurement Science and Technology · doi:10.1088/1361-6501/ad9347 · Cited by 2
“DF-VO [14] bypasses the need to match depth from a single image and camera motion for consecutive images by utilizing optical flow-based feature matching, thus eliminating the need for a pose network.”
Visual and Inertial Sensor Fusion Approach for Visual Inertial Odometry Used in Vehicle Localization
Jo, H., & Yoon, S. W. (2024) · Proceedings of IEEE Sensors/Proceedings of IEEE Sensors ... · doi:10.1109/sensors60989.2024.10785070 · Cited by 2
“In contrast, hybrid deep-geometric methods are proposed in a self-supervised manner [11, 12].”
Rigid-aware self-supervised GAN for camera ego-motion estimation
Lin, L., Luo, W., Yan, Z., & Zhou, W. (2022) · Digital Signal Processing · doi:10.1016/j.dsp.2022.103471 · Cited by 2
Information-Theoretic Odometry Learning
Zhang, S., Zhang, J., & Tao, D. (2022) · International Journal of Computer Vision · doi:10.1007/s11263-022-01659-9 · Cited by 2
“Apart from end-to-end learning, there are also trends in unsupervised learning (Zhou et al., 2017; Yin&Shi, 2018; Ranjan et al., 2019; Bian et al., 2019) and the combination of learned features with geometric methods (Zhan et al., 2020; Yang et al., 2020; Zhang et al., 2022c, b).”
BackgroundUnsupervised Learning of Depth Estimation and Visual Odometry for Sparse Light Field Cameras
Digumarti, S. T., Daniel, J., Ravendran, A., Griffiths, R., & Dansereau, D. G. (2021) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros51168.2021.9636570 · Cited by 2
Monocular weakly supervised depth and pose estimation method based on multi-information fusion
Zhang, Z., Qiao, J., & Lin, S. (2021) · Kuwait Journal of Science · doi:10.48129/kjs.12929 · Cited by 1
Deep-Learning-Based Visual Odometry Models for Mobile Robotics
de Sousa, F. L. M., Meira, N. F. D. C., Oliveira, R., & Silva, M. C. (2021) · doi:10.5753/sbesc_estendido.2021.18504 · Cited by 2
Dense depth estimation from multiple 360-degree images using virtual depth
Yang, S., Kim, K. H., & Lee, Y. (2022) · Applied Intelligence · doi:10.1007/s10489-022-03391-w · Cited by 2
“machine vision applications, including scene reconstruction [8,23,27,30,31,33,34, 36,44], view synthesis [20,26], navigation [42,43,46], and tracking [3, 21].”
BackgroundA Comparison of Deep Learning-Based Monocular Visual Odometry Algorithms
Jeong, E., Lee, J., & Kim, P. (2022) · Lecture notes in electrical engineering · doi:10.1007/978-981-19-2635-8_68 · Cited by 2
Switchable-Encoder-Based Self-Supervised Learning Framework for Monocular Depth and Pose Estimation
Kim, J., Gao, R. Q., Park, J., Yoon, J., & Cho, K. (2023) · Remote Sensing · doi:10.3390/rs15245739 · Cited by 2
Learning Embodied Agents with Policy Gradients to Navigate in Realistic Environments
Staroverov, A., Vetlin, V., Makarenko, S., Naumov, A., & Panov, A. I. (2020) · Studies in computational intelligence · doi:10.1007/978-3-030-60577-3_24 · Cited by 2
Robot Perception enables Complex Navigation Behavior via Self-Supervised LearningPreprint
Chancán, M., & Milford, M. J. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.08967 · Cited by 3
“Only recently, however, the use of more advanced self-supervised learning techniques have enabled to outperform those purely geometry-based or deep-learningbased VO systems [21, 19].”
Semantic and Optical Flow Guided Self-supervised Monocular Depth and Ego-Motion Estimation
Fang, J., & Liu, G. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-87361-5_38 · Cited by 1
AI-Based Visual Odometry Implementation on an Embedded System
Büyüksolak, O., & Güneş, E. O. (2023) · National Conference on Electrical and Electronics Engineering (ELECO) · doi:10.1109/iceee59925.2023.00016 · Cited by 1
“Therefore, while evaluating we realized the 7DoF optimization [22] to obtain the scale that minimizes Absolute Trajectory Error(ATE).”
MethodWang, Z., Shen, M., & Chen, Q. (2024) · Multimedia Tools and Applications · doi:10.1007/s11042-024-18382-x · Cited by 1
“[36]andZhaoetal.”
Occlusion-Aware Monocular Visual Odometry for Robust Trajectory Tracking
Kang, W., Gai, S., Da, F., & Cai, Z. (2025) · IEEE Robotics and Automation Letters · doi:10.1109/lra.2025.3598609 · Cited by 1
Real-Time FPGA-Based Hardware-Algorithm Co-Design for Monocular Visual Odometry on Edge Devices
Huang, L.-Y., Hsieh, Y.-K., & Chien, S. (2026) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2026.3664685 · Cited by 1
“As discussed in Section II-B, DPDM [2] is selected as the baseline over DFVO [19] due to its favorable balance between accuracy and pipeline simplicity.”
A SLAM Pose Graph Optimization Method Using Dual Visual Odometry
Cao, J., Yu, J., Pan, S., Gao, F., Yu, C., Xu, Z., Huang, Z., & Wang, Y. (2021) · Journal of Computer-Aided Design & Computer Graphics · doi:10.3724/sp.j.1089.2021.18663 · Cited by 1
“本文对 Monodepth2 [7] 进行了优化 , 其基于 本文提供优化位姿约束的 VO 是 DFVO [8] 也有选用其他优化约束条件的 , 如 OKVIS [19] 利用惯性传感器作为额外约束 , 将旧状态部分边 缘化 , 同时基于关键帧进行优化对象筛选 , 维护了 有限大小的优化窗口 , 从而确保实时运行 .”
InfluentialDirect RGB-D visual odometry with point features
Yao, Z., An, X., Charrier, C., & Rosenberger, C. (2024) · Intelligent Service Robotics · doi:10.1007/s11370-024-00559-w · Cited by 1
“…addition to the brief review given above, existing studies of VOs have been embracing various possibilities such as joint minimizations considering both geometric and photometric consistencies [9], multiple feature types [27] and learning-based approaches [37] that mainly focused over recent years.”
Enhancing Visual Inertial Odometry Performance using Deep Learning-based Sensor Fusion
Doorshi, R., Saleem, H., & Malekian, R. (2024) · IEEE International Conference on Internet of Things (iThings) and IEEE Green Computing and Communications (GreenCom) and IEEE Cyber, Physical and Social Computing (CPSCom) and IEEE Smart Data (SmartData) · doi:10.1109/ithings-greencom-cpscom-smartdata-cybermatics62450.2024.00040 · Cited by 1
“Huangying Zhan et al.[7] extract IMU and visual features and then utilize the attention mechanism to combine them.”
MethodA Novel LiDAR-Camera Fusion Method for Enhanced Odometry
Dong, H., Xun, Y., He, Y., Liu, J., Mao, B., & Guo, H. (2024) · Globecom · doi:10.1109/globecom52923.2024.10901103 · Cited by 1
“We compare our methods with several state-of-the-art odometry methods, such as classical LO [16], learning-based LO [19], classical VO [20], learning-based VO [21], classical VLO [22], learning-based VLO [23].”
Lu, B., Zhou, T., Li, Q., Chi, W., Wang, Y., Wang, Y., Liu, H., Gu, J., & Sun, L. (2025) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2025.3594113 · Cited by 1
OmniPose6D: Towards Short-Term Object Pose Tracking in Dynamic Scenes from Monocular RGB
Lin, Y., Zhao, Y., Chu, F.-J., Chen, X., Wang, W., Tang, H., Vela, P. A., Feiszli, M., & Liang, K. J. (2025) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros60139.2025.11246949 · Cited by 1
“Metrics: As we work on a novel challenge without a pre-existing, dedicated evaluation metric, we adopt universally recognized evaluation criteria to ensure a fair and accessible benchmark for future work, believing the KITTI Odometry metrics [47] fit our task most.”
FSO-VO: Visual odometry based on dense optical flow prediction and sequence optimization
Liu, G., Zhang, S., Gao, Q., & Yuan, Y. (2026) · Pattern Recognition · doi:10.1016/j.patcog.2026.113434 · Cited by 1
Unsupervised Learning Of Visual Odometry Using Direct Motion Modeling
Andrei, S. S., & Gady Agam, B. (2021) · Proceedings - International Conference on Image Processing · doi:10.1109/icip42928.2021.9506572 · Cited by 1
“The results were copied from [16] and are with loop closure disabled for a more fair comparison.”
BackgroundMethodUsing Self-Supervised Depth and Uncertainty Predictions for Monocular Visual Odometry
Kuang, Y., Huang, P., & Lv, T. (2022) · International Conference on Image and Graphics Processing (ICIGP) · doi:10.1145/3512388.3512435 · Cited by 1
“[15] are chosen as the baseline and our model is compared with the state-of-the-art methods, such as [8], [17] and [19].”
BackgroundMethodInfluentialMOLTR: Multiple Object Localisation, Tracking, and Reconstruction from Monocular RGB VideosPreprint
Li, K., Rezatofighi, H., & Reid, I. D. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2012.05360 · Cited by 1
“To this end, we demonstrate that our system can work with estimated camera trajectory from DF-VO [46] on KITTI dataset.”
BackgroundMethodInfluentialSelf-citationCoFi: Coarse-to-Fine ICP for LiDAR Localization in an Efficient Long-lasting Point Cloud MapPreprint
Lyu, Y., Huang, X., & Zhang, Z. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.10194 · Cited by 1
“Seq [33] [71] [26] [72] LDSO [73] [74] [75] CNN-SVO [76] [77] DDSO [78] DSO [79] ORB-SLAM2 [80] CoFi (ours) 00 0.”
Deep Inertial Odometry with Accurate IMU PreintegrationPreprint
Khorrambakht, R., Lu, C. X., Damirchi, H., Chen, Z., & Li, Z. G. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2101.07061
“We use the open-source implementation provided in [13].”
Fang, J., & Guizhong, L. (2021) · IEEE International Conference on Unmanned Systems (ICUS) · doi:10.1109/icus52573.2021.9641225
“For the KITTI odometry dataset, the sequences 00-08 was used for training and 09-10 for testing [39],[40].”
MethodParameter-adaptive multi-frame joint pose optimization method
Li, S., Xian, Y., Wu, W., Zhang, T., & Li, B. (2022) · The Visual Computer · doi:10.1007/s00371-022-02476-4
LiDAR-Camera Fusion for Depth Enhanced Unsupervised Odometry
Fetic, N., Aydemir, E., & Ünel, M. (2022) · IEEE Vehicular Technology Conference · doi:10.1109/vtc2022-spring54318.2022.9860840
“and Lfc is a bidirectional census loss computed using backward and forward flow as in [3].”
BackgroundMethodInfluentialSelf-Supervised Ground-Relative Pose Estimation
Muller, B. R., & Smith, W. A. P. (2022) · Proceedings - International Conference on Pattern Recognition/Proceedings/International Conference on Pattern Recognition · doi:10.1109/icpr56361.2022.9956720
RGB-D Visual Odometry via Low Order Gaussian Gradient MetricsPreprint
Yao, Z., An, X., Charrier, C., & Rosenberger, C. (2023) · doi:10.36227/techrxiv.22132544.v1
RGB-D Visual Odometry via Low Order Gaussian Gradient MetricsPreprint
Yao, Z., An, X., Charrier, C., & Rosenberger, C. (2023) · doi:10.36227/techrxiv.22132544
SMCEWS: Binary Robust Multicentre Features
Xu, Y., Chen, M., Zhang, W., Li, H., Yang, R., & Wang, Y. (2023) · Symmetry · doi:10.3390/sym15040809
Researches Advanced in Visual Simultaneously Localization and Mapping based on Deep Learning
Hao, H. (2022) · International Conference on Robotics, Intelligent Control and Artificial Intelligence (RICAI) · doi:10.1145/3584376.3584595 · Cited by 1
“[15] using two CNNs for depth estimation and pose prediction simultaneously both have reduced the displacement error in the KITTI dataset to about 4%.”
MethodBilal, M., Hanif, M. S., Munawar, K., & Al-Saggaf, U. M. (2023) · IEEE Access · doi:10.1109/access.2023.3284463
RGB-D Visual Odometry via Low Order Gaussian Gradient MetricsPreprint
Yao, Z., An, X., Charrier, C., & Rosenberger, C. (2023) · doi:10.36227/techrxiv.22132544.v2
Motion Consistency Loss for Monocular Visual Odometry with Attention-Based Deep Learning
Françani, A. O., & Máximo, M. R. O. A. (2023) · Latin American Robotics Symposium, Brazilian Symposium on Robotics and Workshop on Robotics in Education (LARS/SBR/WRE) · doi:10.1109/lars/sbr/wre59448.2023.10332921 · Cited by 1
“Prior works evaluate monocular methods with an optimization transformation to align the predictions with the ground truths [10], [13].”
BackgroundMethodInfluentialMulti-Masks Generation for Increasing Robustness of Dense Direct Methods
Liu, Z., Malis, E., & Martinet, P. (2023) · IEEE International Conference on Intelligent Transportation Systems (ITSC) · doi:10.1109/itsc57777.2023.10422213
Hybrid Visual Odometry Algorithm Using a Downward-Facing Monocular Camera
Al‐Hadithi, B. M., Thomas, D., & Pastor, C. (2024) · Applied Sciences · doi:10.3390/app14177732
“These meth-ods have shown great performance both in the development of VO systems [8,14,19–21] or just in improving the specific task of geometry-based algorithms, such as aiding with depth estimation in monocular cameras [22] or using the concept of aleatory uncertainty to calculate photometric…”
Towards Accurate Ego-lane Identification with Early Time Series Classification
Jin, Y., Stenhammar, T., Bejmer, D., Beauvisage, A., Xia, Y., & Fu, J. (2024) · International Conference on Information Fusion (FUSION) · doi:10.23919/fusion59988.2024.10706312
“Over the past decades, a range of algorithms, including e.g., INS/GNSS navigation [1]–[3], Visual Odometry (VO) [4], [5], Simultaneous Localization and Mapping (SLAM) [6], [7], and Visual-Inertial Navigation Systems (VINS) [8], have been developed to address the challenges in localization.”
BackgroundChen, S., Kong, D., Lu, L., Chen, D., & Chen, D. (2024) · Measurement Science and Technology · doi:10.1088/1361-6501/ad976b
“To further illustrate the performance of the DyPanVO, we additionally introduce DAVO [27], CM-VO [41], NeuralBundler [42], and the hybrid algorithm DF-VO [43] for quantitative comparison.”
Eum, S., Lee, H., Kwon, H., Osteen, P., & Harrison, A. (2024) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros58592.2024.10801483
“To avoid such inconvenience, attempts have been made to leverage depth only in training while addressing tasks such as 3D scene reconstruction [22], [23], visual odometry [24], [25], [26], [27], monocular 3D object detection [28], [11], panoptic segmentation [29], [30], frame interpolation [31].”
Research on Stereo Vision Localization Method Based on Prior Semantic Map
Ma, X., Shi, Y., Lai, Z., & Fang, H. (2024) · International Conference on Robotics, Intelligent Control and Artificial Intelligence (RICAI) · doi:10.1109/ricai64321.2024.10911237
“However, in practical applications, visual odometry tends to accumulate errors [3].”
SIE-DepthNet: Semantic-Guided Monocular Depth Estimation for Dynamic Environment
Song, Z., Gao, Y., Dai, S., Li, S., Hao, A., & Zhang, S. (2025) · Lecture notes in computer science · doi:10.1007/978-981-96-3679-2_13
Visual Slam in the Heterogeneous Intelligence Era: A SurveyPreprint
Su, S., He, X., Meng, X., Zhang, Y., Yin, W., Mo, L., & Yu, F. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5383002
SATVIO: Stereo Attention-Based Visual Inertial Odometry
Doorshi, R., Saleem, H., & Malekian, R. (2025) · IEEE Journal of Selected Areas in Sensors · doi:10.1109/jsas.2025.3601056
“In addition, the integration of attention mechanisms, such as those proposed by Zhan et al. [8], Tu et al. [14], and Wang et al. [15], highlights a significant trend toward making these systems more adaptive to the nuances of real-world environments.”
The Autonomy Loop: Today’s Practices and Tomorrow’s Challenges
Speranzon, A., Varadarajan, S., Yenawine, A., Ramsey, J. A., DeGalan, L. H., Grebe, G. S., Fregene, K., Kwon, D. W., Niestroy, M. A., Franke, J. L., et al. (2025) · Proceedings of the ... American Control Conference/Proceedings of the American Control Conference · doi:10.23919/acc63710.2025.11107541
“Recently, visual odometry has leveraged deep learning approaches [66], combining deep networks to estimate depth and optical flow and combining this information using epipolar geometry.”
Self-Supervised Monocular Depth Estimation with Dual Mask and Attention Mechanism
Jiang, R., Gai, S., & Da, F. (2025) · Chinese Control Conference (CCC) · doi:10.23919/ccc64809.2025.11179816
“Since monocular visual odometry is unable to recover scale without auxiliary inputs, we align all result with ground truth trajectories in 7dof mode in DFVO[41].”
Self-Supervised Geometry-Guided Initialization for Robust Monocular Visual Odometry
Kanai, T., Vasiljevic, I., Guizilini, V., & Shintani, K. (2025) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros60139.2025.11247022 · Cited by 1
“• We provide a comprehensive analysis of our proposal adapted to DROID-SLAM [1] on the standard KITTI benchmark [15], as well as challenging driving scenes from DDAD [14], and provide insights into how to construct future visual SLAM systems.”
A visual localization approach based on the fusion of semantic depth and inertial information
Zhou, M., Zhang, Y., Wu, D., Feng, Y., & Qu, S. (2025) · doi:10.1117/12.3095716
“For example, Zhan et al. 3 combined CNNs with geometric constraints to estimate depth maps and optical flow, while Zhang et al. 4 introduced SfM-Learner, an unsupervised framework for jointly learning scene depth and camera pose from video sequences.”
Long-Term Feature Point Tracking for Camera Pose Estimation in Forest Fire Scenes
Zhang, H., & Lu, X. (2025) · IEEE International Conference on Big Data and Artificial Intelligence (BDAI) · doi:10.1109/bdai66031.2025.11325611
“Notably, VO systems that depend on dense optical flow for pose estimation [13], [23], [24] are particularly prone to failure if the assumption of photometric consistency is violated, leading to significant challenges in accurate pose prediction.”
Dual Attention XYDeblur: Attention-Guided Divide-and-Conquer Strategy for Image Deblurring
Shin, J. S., & Lee, H. S. (2026) · IEEE Access · doi:10.1109/access.2026.3660952
“…of this manuscript and approving it for publication was Patrizia Livreri . as a crucial preprocessing step for various computer vision applications, including object detection [1], [2], semantic segmentation [3], [4], visual odometry [5], [6], and planning-oriented autonomous driving [7], [8].”
Choi, J. H., Kang, C. H., & Kim, S. Y. (2026) · International Journal of Control Automation and Systems · doi:10.1007/s12555-026-00029-2
Niu, Q. (2025) · International Conference on Mechanical Automation and Electronic Information Engineering (MAEIE) · doi:10.1109/maeie68099.2025.11406022
“Zhan et al [27]. proposed DeepVO, the first end-to-end visual odometry system, which uses a convolutional-recurrent network architecture to directly regress 6-DoF poses from image sequences.”
Detection and Tracking of Mesh Intersection Points for Autonomous Net Cleaning Robots
Li, G., Wang, J., Lian, A., Gou, L., Pang, G., Yuan, T., Hu, Y., & Huang, X. (2026) · Fishes · doi:10.3390/fishes11040215
Contributions to Plane-based and Deformable 3D Reconstruction from Monocular Images
Collins, T. (2021) · theses.fr (ABES) · doi:10.70675/82e2886fz5a90z4f52zb0efz1b215c25f94c
Yuan, Z., Shen, Y., Wang, J., & Wang, Y. (2026) · Research Square · doi:10.21203/rs.3.rs-9365602/v1
Self-Supervised Monocular Depth Estimation With Multimodal Fusion and Pose Enhancement
Chen, H., Wang, Y., Jia, S., Huang, S., & Zhang, J. (2026) · IEEE Sensors Journal · doi:10.1109/jsen.2026.3684347
Detecting gaze shifts of moving observers in dynamic environments
Nejad, A., Ghiani, A., Cornelissen, F. W., & Brenner, E. (2026) · Behavior Research Methods · doi:10.3758/s13428-026-02972-8
“…in a head-centered reference frame, measures scene content similarity using image patches around the gaze point using 2ch2stream neural network (Zagoruyko & Komodakis, 2015), and estimates head and body motion using a visual odometry method, known as DF-VO (Zhan, Weerasekera, Bian, & Reid, 2020).”
Self-supervised Monocular Visual Odometry with Scale-Consistency
Yan, X., CHEN, W., Tong, B., Si, F., Du, L., & Fan, J. (2025) · China Automation Congress (CAC) · doi:10.1109/cac67268.2025.11487862
“Nevertheless, learning-based monocular VO methods still face challenges in achieving accurate and globally consistent pose estimation due to inherent scale ambiguity and the sensitivity of photometric losses in textureless or non-Lambertian regions [7].”
Relative Pose Estimation Using Optical Flow-Guided Hybrid Supervision for Visual Odometry
Chen, X., Cao, Y., & Yu, H. (2026) · IEEE Access · doi:10.1109/access.2026.3688850
“On KITTI, we categorize the related work into three groups: geometric methods [1], [43], [44], as well as self-supervised [24], [25], [26], [34], [35], [36] and supervised [6], [19], [27], [28], [29], [45], [46], [47] in order to clearly distinguish different research paradigms.”
FOCUS: A Failure-Tolerant Visual-Inertial Odometry Using Contextual Understanding and Sensor Fusion
Nazeer, N., Leier, M., Reinsalu, U., & Tepljakov, A. (2026) · International Conference on Mechatronics and Robotics Engineering (ICMRE) · doi:10.1109/icmre69538.2026.11533894
“Furthermore in a challenging sequence such as 01 (as agreed by the authors in [11]), our proposed system performs very well and requires no prior map re-initialization and does not fail mid-sequence.”
InfluentialSA4Depth: Consistent Pose-Depth Scale Alignment for Self-Supervised Monocular Depth Estimation
Li, C., Berner, N., Navab, N., Tombari, F., & Gasperini, S. (2026) · IEEE Robotics and Automation Letters · doi:10.1109/lra.2026.3699248
“The average errors in translation and rotation are calculated following [38].”
Combining Projected Uncertainty for Self-Supervised Visual Odometry: From Two-Frame to Multi-Frame
Xie, J., Dhaouadi, O., Chen, W., Meier, J., Bauer, Z., Pollefeys, M., & Cremers, D. (2026) · International Journal of Computer Vision · doi:10.1007/s11263-026-02915-y
“01, DFVO (Zhan et al., 2020) diverges from ground truth, while SC-Depth (Bian et al., 2021) exhibits substantial drift.”
InfluentialTong, B., Chen, W., Du, L., Wang, H., Yan, X., & Si, F. (2026) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2026.3711340
“The comparison baselines encompass geometry-based monocular VO [7], end-to-end learning-based approaches [9], [20], [22], [48], and hybrid frameworks that integrate deep learning with geometric constraints [23], [24], [25], [27], [29].”
GeM-VO: Deep Stereo Odometry With Geometry Masks in Dynamic Environments
Xie, C., Yao, H., Xing, R., Hao, N., & He, F. (2026) · IEEE Sensors Journal · doi:10.1109/jsen.2026.3712373
“More recently, with the advancement of deep learning techniques, end-to-end methods have attracted growing attention [6], [7].”
Eye-Tracking Using Web-Camera and Deep Learning
Ligostaev, N., Orlova, J., Stepanov, A., & Somov, A. (2026) · Signals and communication technology · doi:10.1007/978-3-032-18566-2_15
GCAFormer-VO: A Geometry and Correspondence-Aware Transformer for Visual Odometry
Bao, J., Wu, Q., Huang, J., & Yuan, X. (2026) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2026.3735107
“Moreover, DF-VO [25] bridges learned representations with classical geometric optimization by training deep features that better support geometric estimation, retaining an optimization-based structure while benefiting from learned robustness.”
Transformer-Based End-to-End Model for Monocular Visual Odometry on Urban Traffic Scenes
Pontarolo, G., & Todt, E. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-39898-7_33
Transformer-Based Model for Monocular Visual Odometry: A Video Understanding Approach
Françani, A. O., & Maximo, M. R. O. A. (2023) · arXiv · doi:10.48550/arxiv.2305.06121 · Cited by 38
“Among the approaches to address the monocular VO, there are traditional geometry-based methods [2, 3], deep learning-based methods that use end-to-end architectures [4, 5], and hybrid methods with deep learning in certain modules of the geometry-based methods [6, 7, 8].”
Metadata
- Authors
- Zhan, H., Weerasekera, C. S., Bian, J.-W., & Reid, I.
- Affiliations
- The University of Adelaide (AU)
- Venue
- 2020 IEEE International Conference on Robotics and Automation (ICRA)
- Publisher
- IEEE
- Published
- 2020-05
- Pages
- 4203-4210
- ISSN
- 1050-4729 (ISSN-L)
- OpenAlex
- W3091687063
- Cited by
- 201
- References
- 66
- Citations by year
- 2026: 18 · 2025: 28 · 2024: 37 · 2023: 37 · 2022: 33 · 2021: 30 · 2020: 17
- FWCI
- 43.12
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- Locations
- Proceedings - IEEE International Conference on Robotics and Automation/Proceedings (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- Adelaide Research & Scholarship (AR&S) (University of Adelaide) (submittedVersion)
- arXiv (Cornell University)
- Funders
- Australian Centre for Robotic Vision
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Robotics and Sensor-Based Localization
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Vision and Imaging
- Physical Sciences › Earth and Planetary Sciences › Geology › 3D Surveying and Cultural Heritage
- Keywords
- monocular visual odometry, scale drift, single-view depth estimation, optical flow, epipolar geometry, perspective-n-point, convolutional neural networks
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
