Pith. sign in

REVIEW 3 major objections 6 minor 41 references

Robust Video-Based Pothole Detection and Area Estimation for Intelligent Vehicles with Depth Map and Kalman Smoothing

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read A video-only pipeline estimates pothole area from depth maps and Kalman smoothing, replacing flat-road geometric priors with learned metric depth.

desk verdict A plausible detection++ pipeline whose headline area-estimation claim is untestable as reported because it is validated only for self-consistency, not against measured ground truth. read the letter →

arxiv 2505.21049 v1 pith:WMMKWULB submitted 2025-05-27 cs.CV

classification cs.CV
keywords potholedetectionareaestimationmonocularmetricdepthminimumboundingtriangulatedpixelKalmanfiltersmoothingsmallobjectvideotrackingautonomousdriving
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to show that pothole area can be estimated reliably from a single vehicle-mounted camera, without LiDAR or a flat-road geometric prior. It builds a video pipeline that detects potholes, tracks them across frames, and converts each bounding box into 3D measurements using per-pixel monocular depth. The two new components are the Minimum Bounding Triangulated Pixel (MBTP) method, which sums triangular facets over the pothole's projected bounding region, and the Confidence-and-Distance Kalman Filter (CDKF), which smooths area estimates using detection confidence and distance. The authors report that MBTP reduces estimation fluctuation relative to a corner-point baseline and that CDKF further stabilizes consecutive-frame estimates on clear and dark road datasets. If correct, the framework offers a low-cost complement to LiDAR-based road condition monitoring.

What carries the argument

The named central object is the Minimum Bounding Triangulated Pixel (MBTP) method: it projects each pixel in a detected bounding box into camera coordinates via the pinhole model, forms a 2×2-pixel quadrilateral in 3D, splits it into two triangles, and sums the triangle areas, then scales by π/4 on the assumption that potholes are roughly elliptical. The smoothing mechanism is the Confidence-and-Distance Kalman Filter (CDKF), which keeps a constant-state model of pothole area per tracked identity and sets the measurement noise R such that close, confident detections are trusted more than distant, uncertain ones.

What would settle it

Measure real pothole areas on a held-out set using a tape measure or LiDAR and compare them with MBTP estimates; if the error grows systematically with distance or shows a roughly constant multiplicative bias, the depth-scale or camera-intrinsics assumption is wrong even when frame-to-frame consistency looks good.

Watch

Extended reading notes

Core claim

The paper's central claim is that area estimates from monocular video become both accurate and stable when you combine a small-object-sensitive detector with per-pixel metric depth and a Kalman smoother whose measurement noise depends on detection confidence and distance. On the two test datasets, the proposed detector reaches an AP(50) of 76.6% under clear conditions and 72.2% under dark conditions, and the MBTP area estimator outperforms the corner-point method on mean absolute error, coefficient of variation, and adjacent-frame differences; adding CDKF lowers the best adjacent-frame difference to 0.02 m² (clear) and 0.009 m² (dark). The broader message is that geometric assumptions like flat road surfaces can be replaced by learned depth, and that video consistency can be enforced with a lightweight filter.

Load-bearing premise

The absolute area numbers rest on the pre-trained depth model returning true metric depth that matches the camera's focal length and optical center, but the paper does not verify the intrinsics or the depth scale on the test data.

Editorial extensions

If this is right

  • Area estimation no longer requires flat-road assumptions or known distance priors; any frame with a detection and a depth map yields an area estimate.
  • Small and distant potholes, the hardest cases, are where the added small-object detection head contributes the largest gains in recall and precision.
  • Smoothing across tracked frames makes per-frame estimates less noisy, so a vehicle system can act on measurements without re-deriving a filter for each lighting condition.
  • The reported per-frame runtime of roughly 110 ms with parallel processing is compatible with near-real-time in-vehicle deployment.
  • The same detection-plus-depth-plus-tessellation recipe transfers to other ground-surface defects, such as cracks, patches, and manhole covers, with only detector retraining.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the paper reports no camera intrinsics and no depth-scale validation, the absolute area values are the least certain part; a uniform rescale factor could correct them if ground-truth areas became available, provided the depth scale is spatially consistent.
  • The π/4 ellipse factor inside MBTP imposes a fixed-shape bias; replacing it with actual instance masks should improve accuracy for non-elliptical potholes.
  • A testable extension is to run the same pipeline on a set with measured pothole areas, converting the consistency metrics into absolute accuracy numbers and recalibrating CDKF's noise weights for that sensor.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes a video-based framework for detecting potholes and estimating their surface area from a monocular camera. The pipeline consists of: (i) an object detector, ACSH-YOLOv8, which adds a P2 small-object head and an ACmix attention module to YOLOv8; (ii) BoT-SORT with ego-motion compensation for multi-frame tracking; (iii) metric depth maps generated by DepthAnything V2; (iv) a new Minimum Bounding Triangulated Pixel (MBTP) area estimator that back-projects the detection bounding box to 3D using a pinhole model and sums triangular facets; and (v) a Kalman filter variant, CDKF, whose measurement noise is set from detection confidence and pothole distance. Experiments are conducted on the Bučko Clear and Dark Road datasets. Detection results are compared against YOLO baselines using AP and related metrics. Area estimation results are evaluated with MAE, CV, AFD, and NIS, with and without Kalman smoothing, and with different choices of noise covariance. The manuscript claims that MBTP outperforms a corner-point baseline and that CDKF improves temporal consistency.

Significance. If the area-estimation claims were convincingly validated, the work would be a useful engineering contribution to vision-based road-survey systems: it combines standard components in a sensible pipeline, and the detection module is assessed against ground-truth bounding boxes with a clear AP improvement. The runtime analysis (Table III) is also a positive feature, showing practical per-frame timings. However, the central claim of the paper is accurate and robust pothole area estimation, and this claim is not supported by the evidence as presented. The evaluation of area estimates uses only internal consistency metrics, never comparisons against physically measured areas, and a direct contradiction appears between the stated absence of ground truth and a later claim of agreement with ground truth. Furthermore, the Kalman filter parameters are tuned on the same evaluation data with an objective that is a weighted sum of the very metrics reported as improvements, so the reported gains are partly in-sample fits. These issues undermine the key contribution of the manuscript.

major comments (3)
  1. [Section IV.C.1 and Section IV.E] The evaluation protocol for area estimation is internally contradictory and cannot support the claimed accuracy. Section IV.C.1 states, 'Given the absence of ground truth measurements, we evaluate the estimation method's accuracy and consistency using statistical measures based on multiple observations of the same pothole,' and Eqs. (21)-(24) define only self-consistency metrics. Yet in Section IV.E, the paragraph following Fig. 11 asserts that MBTP estimates 'align well with ground truth measurements' and that the CP method 'overestimates' relative to a 1 m² reference. No ground-truth measurement protocol, numerical comparison, or error statistic is provided. The central claim of accurate area estimation is therefore untestable from the reported data; either the ground-truth comparison exists and must be presented, or the claim of alignment should be removed.
  2. [Section IV.C.2 and Table II] The CDKF improvements are obtained by optimizing λ and θ on the same evaluation data used to report Table II, making the reported gains partly a result of fitting rather than predictive robustness. Equation (27) minimizes J(λ, θ) = 10·MAE + CV + AFD + NIS, where these four metrics are exactly the ones reported in Table II. The Bayesian optimization is run on the evaluation sequences (four trials of initialization plus 30 iterations), and the best parameters are then used to produce Table II. This is a circular evaluation: the filter is tuned to minimize the outcome metrics on the test data, so the lower MAE, CV, AFD, and NIS values for the CDKF rows do not demonstrate a generalizable improvement. The authors should either use a held-out validation set for parameter selection or report the sensitivity of the results to λ and θ over a plausible range without in-sample tuning.
  3. [Section III.C and III.D (Eqs. 6-7)] The absolute area estimates are not traceable to any validated metric scale. Equations (6)-(7) require the camera intrinsics (pu, pv, fu, fv), but these are not reported for the Bučko datasets. Additionally, DepthAnything V2 is pre-trained on KITTI, and the paper does not validate whether its output is metric depth directly usable with the pinhole model on the Bučko camera, nor does it provide any check of depth scale on the test data (e.g., comparing to known road markings or measured distances). If the depth scale or intrinsics are wrong, all reported areas are multiplied by an unknown constant or spatially varying factor, making the claimed '0.2 m²' average and the comparison between methods unverifiable. The authors should report the intrinsics and provide a depth-accuracy validation on at least a few frames.
minor comments (6)
  1. [Section III.E] The acronym CDKF is defined as 'Kalman Filter based on Confidence and Distance' in the contribution list, but the section heading and later text say 'Confidence Distance.' Please use one consistent name.
  2. [Equation (1)] The notation Nk(i,j) for the local pixel region is not clearly defined, and the equation is typeset in a way that makes the softmax argument hard to parse. Please rewrite the equation with explicit indexing and define all symbols.
  3. [Section II.B] The term 'UAV-Structure-from-Motion' should be 'UAV Structure-from-Motion' (no hyphen after UAV).
  4. [Conclusion (Section V)] In the first paragraph, 'prpposed' is a typo for 'proposed'.
  5. [Table II] Table II is very hard to read because the row labels are abbreviated with quotation marks and the structure is unclear. Each row should independently state the method, e.g., 'CP', 'MBTP', 'MBTP + KF (confidence only)', 'MBTP + KF (distance only)', 'MBTP + CDKF (combined)'. In the current format, a reader cannot tell which rows correspond to which configuration.
  6. [Section IV.C.2] The Python library is written as 'bayes opt'; the correct package name is 'bayes_opt'. Please also state the acquisition function and the number of data splits used, so the optimization setup is reproducible.

Circularity Check

2 steps flagged · score 7.0 of 10

CDKF 'robustness' gains are fitted to the same metrics reported as improvements, while area 'accuracy' is measured by internal consistency rather than ground truth, making the central area-estimation claim partially circular.

  1. fitted input called prediction [Section IV.C.2, Eqs. (26)-(27); Section V.E, Table II discussion]
    "To determine the optimal weights lambda and theta for these two noise factors, we employ a Bayesian optimization algorithm to maximize overall filtering performance. J(λ, θ) = 10·MAE+CV+AFD+NIS (26) ... (λ∗, θ∗) = arg minJ(λ, θ)(27) ... The weights, λ and θ, are optimized using Bayesian optimization. ... The results ... are shown in the lower section of Table. II."

    Eq. (15) makes R a function of λ and θ, and Eq. (26) defines the reported evaluation metrics (MAE, CV, AFD, NIS) as the objective J. Eq. (27) then selects λ and θ on the same evaluation sequences used to produce Table II. The CDKF row of Table II is therefore not an independent prediction: it is the post-fit value of the very objective that was minimized. The reported AFD, MAE, CV, and NIS gains reduce to parameter fitting on the evaluation data, so the paper's claim that CDKF 'optimization' makes estimates 'more robust' is circular evidence.

  2. self definitional [Section III.E, Eq. (16); Section IV.C.1, Eqs. (21)-(24); Section V.E]
    "For an individual pothole, it is assumed that its area remains constant across consecutive frames... Ak = A k|k-1+K k-1(zk-A k|k-1)(16)... Given the absence of ground truth measurements, we evaluate the estimation method's accuracy and consistency using statistical measures based on multiple observations of the same pothole."

    The CDKF output is a weighted average of the previous estimate and the current measurement, so a Kalman smoother is designed by construction to damp frame-to-frame variation. The paper's own metrics quantify exactly that property: AFD is the average adjacent-frame change, while MAE and CV measure spread around the mean of the same output sequence. Reporting lower MAE/CV/AFD after smoothing is therefore reporting the filter's design objective rather than an independent accuracy result. This is compounded by the contradiction that Section IV.C.1 declares 'absence of ground truth measurements,' while Section V.E asserts that the estimates 'align well with ground truth measurements' without supplying any independent area measurement protocol.

full rationale

The detection module (ACSH-YOLOv8) is validated against annotated bounding boxes with AP, recall, and precision, so that part of the paper is not circular. The circularity is concentrated in the area-estimation and CDKF claims. First, the CDKF hyperparameters λ and θ are chosen by Bayesian optimization over the objective J = 10·MAE+CV+AFD+NIS (Eqs. 26-27), and Table II then reports exactly those quantities as evidence of improvement; because Eq. 15 makes R a function of λ and θ, the CDKF row is a post-fit description of the tuned filter, not an out-of-sample result. Second, the area 'accuracy' metrics are internal consistency measures computed from repeated estimates of the same pothole, with no comparison to physically measured area. A Kalman smoother is explicitly designed to stabilize such estimates, so the reported reduction in MAE/CV/AFD is partly definitional. The paper even states in Section IV.C.1 that this is due to 'the absence of ground truth measurements,' yet Section V.E claims the estimates 'align well with ground truth measurements,' an unsupported contradiction. The only area-estimation baseline is the authors' own prior Corner Point method (reference [41]), which is a self-citation; it is an executable baseline and not by itself the derivation-circular step, but it reinforces the lack of independent validation. Overall, the detection result is non-circular while the central accuracy and robustness claims for area estimation are partially circular, meriting a score of 7.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The framework depends on several unverified assumptions: known camera intrinsics, metric depth from a pretrained model without domain adaptation, an elliptical pothole shape, constant area over time, and an ad hoc noise model for the Kalman filter. The two Kalman tuning parameters and several other hyperparameters are fitted or unspecified, so the absolute area estimates are not self-contained.

free parameters (4)
  • CDKF lambda (lambda) = Clear: 1.026, Dark: 1.51
    Weight balancing confidence and distance in R (Eq. 15), found by Bayesian optimization of J (Eq. 26-27) on the evaluation data itself.
  • CDKF theta (theta) = Clear: 0.7179, Dark: 1.227
    Tuning factor for the distance term in R (Eq. 15), optimized on the same data used to report the results.
  • Trusted distance d0 = not specified
    Threshold inside which distance no longer affects R (Eq. 15); its value is never given in the paper.
  • Process noise Q and initial covariance P = not specified
    Kalman filter parameters in Eq. 13; their values are not reported, leaving the filter behavior incompletely specified.
assumptions (6)
  • domain assumption Pinhole camera model with known intrinsic parameters (f_u, f_v, p_u, p_v)
    Eqs. 6-7 require camera intrinsics, which are never provided for the Bucko dataset; without them the 3D projection and absolute area are undefined.
  • domain assumption DepthAnything V2 produces metric depth consistent with the pinhole model across both datasets
    The pipeline assumes the pre-trained model outputs absolute depth in the same units as the camera coordinates, with no domain adaptation or scale alignment described (Section III.C).
  • domain assumption Potholes are elliptical, so area equals rectangle area times pi/4
    Eq. 12 applies pi/4 to the sum of triangles over the bounding rectangle, assuming an ellipse inscribed in the rectangle; actual pothole shapes vary widely, and the bounding box also contains non-pothole pixels.
  • domain assumption A pothole's physical area is constant across video frames
    The CDKF constant-state model (Eq. 13) assumes the area does not change over time, which is reasonable for short clips but excludes cases where the pothole is partially occluded or the view changes the visible portion.
  • ad hoc to paper Detection confidence and distance to camera together determine measurement noise R as in Eq. 15
    This functional form (R = lambda / (c + theta*max(d,d0))) is introduced specifically for this paper and is not derived from measurement statistics.
  • domain assumption The minimum bounding rectangle of projected bounding-box pixels represents the pothole footprint
    MBTP sums areas over the entire detection rectangle, including background pixels around the pothole; the method does not segment pothole pixels from the bounding box.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Robust Video-Based Pothole Detection and Area Estimation for Intelligent Vehicles with Depth Map and Kalman Smoothing." pith.science (2026). https://pith.science/paper/WMMKWULB

@misc{pith2026250521049,
  author       = {Pith},
  title        = {Pith review of: Robust Video-Based Pothole Detection and Area Estimation for Intelligent Vehicles with Depth Map and Kalman Smoothing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WMMKWULB}},
  note         = {Machine review of arXiv:2505.21049}
}
read the original abstract

Road potholes pose a serious threat to driving safety and comfort, making their detection and assessment a critical task in fields such as autonomous driving. When driving vehicles, the operators usually avoid large potholes and approach smaller ones at reduced speeds to ensure safety. Therefore, accurately estimating pothole area is of vital importance. Most existing vision-based methods rely on distance priors to construct geometric models. However, their performance is susceptible to variations in camera angles and typically relies on the assumption of a flat road surface, potentially leading to significant errors in complex real-world environments. To address these problems, a robust pothole area estimation framework that integrates object detection and monocular depth estimation in a video stream is proposed in this paper. First, to enhance pothole feature extraction and improve the detection of small potholes, ACSH-YOLOv8 is proposed with ACmix module and the small object detection head. Then, the BoT-SORT algorithm is utilized for pothole tracking, while DepthAnything V2 generates depth maps for each frame. With the obtained depth maps and potholes labels, a novel Minimum Bounding Triangulated Pixel (MBTP) method is proposed for pothole area estimation. Finally, Kalman Filter based on Confidence and Distance (CDKF) is developed to maintain consistency of estimation results across consecutive frames. The results show that ACSH-YOLOv8 model achieves an AP(50) of 76.6%, representing a 7.6% improvement over YOLOv8. Through CDKF optimization across consecutive frames, pothole predictions become more robust, thereby enhancing the method's practical applicability.

Figures

Figures reproduced from arXiv: 2505.21049 by the authors.

Figure 1
Figure 1. Overall flowchart of the proposed pothole area estimation model. The Proposed ACSH-YOLOv8 model is trainable, while DepthAnything V2 uses a [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The ACSH-YOLOv8 model is used for pothole detection, incorporating the P2 detection head and the ACmix hybrid attention mechanism, which are [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Schematic diagram of the ACmix hybrid attention mechanism. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Schematic diagram of the monocular depth model, DepthAnything V2, used to obtain the depth values of pothole pixels. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Schematic diagram of the proposed MBTP method. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Distribution of object center and size in Clear Road Dataset. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Distribution of object center and size in Dark Road Dataset. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Comparison of pothole area estimation workflows under Clear Road Dataset. The first, second and third rows show pothole detection results using [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: Comparison of pothole area estimation workflows under Dark Road Dataset. The first, second and third rows show pothole detection results using [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Fluctuation comparison of pothole area estimation using different [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]
Figure 11
Figure 11. Figure 11: Visualization of area estimation fluctuations for the same pothole across consecutive frames using the different area estimation method and optimization [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

41 extracted references · 35 canonical work pages

  1. [1]

    Aidi Wang, Hong Lang, Zhen Chen, Yichuan Peng, Shuo Ding, and Jian John Lu. The two-step method of pavement pothole and raveling detection and segmentation based on deep learning.IEEE Transactions on Intelligent Transportation Systems, 25(6):5402–5417, 2024

  2. [2]

    Nur Ridzuan Bukhari and Mohamad Yusri Aman. Review study of identify main factor that causes the cracking and potholes on asphalt pavement in malaysia.Recent Trends in Civil Engineering and Built Environment, 4(2):338–350, 2023

  3. [3]

    Pothole-and patch repair failure recurrence in gauteng: The human influence

    J Muller and A Marnewick. Pothole-and patch repair failure recurrence in gauteng: The human influence. In2016 IEEE International Confer- ence on Industrial Engineering and Engineering Management (IEEM), pages 243–247. IEEE, 2016

  4. [4]

    Antarctica warming up: Scientists uncover climate secrets, November 2023

    BBC News. Antarctica warming up: Scientists uncover climate secrets, November 2023. Accessed: 2024-11-14

  5. [5]

    Potholes & more potholes: is it just us?, 2018

    Rick Snyder. Potholes & more potholes: is it just us?, 2018. Accessed: 2024-11-20

  6. [6]

    Effects of pothole type pavement distress on whole-body vibration.Road Materials and Pavement Design, 24(6):1403–1424, 2023

    Ufuk Kırbas ¸. Effects of pothole type pavement distress on whole-body vibration.Road Materials and Pavement Design, 24(6):1403–1424, 2023

  7. [7]

    Computer vision for road imaging and pothole detection: a state-of-the-art review of systems and algorithms.Transportation safety and Environment, 4(4):tdac026, 2022

    Nachuan Ma, Jiahe Fan, Wenshuo Wang, Jin Wu, Yu Jiang, Lihua Xie, and Rui Fan. Computer vision for road imaging and pothole detection: a state-of-the-art review of systems and algorithms.Transportation safety and Environment, 4(4):tdac026, 2022

  8. [8]

    Pothole detection using computer vision and learning.IEEE Transactions on Intelligent Transportation Systems, 21(8):3536–3550, 2019

    Amita Dhiman and Reinhard Klette. Pothole detection using computer vision and learning.IEEE Transactions on Intelligent Transportation Systems, 21(8):3536–3550, 2019

Show all 41 references
  1. [9]

    Road pothole extraction and safety evaluation by integration of point cloud and images derived from mobile mapping sensors.Advanced Engineering Informatics, 42:100936, 2019

    Hangbin Wu, Lianbi Yao, Zeran Xu, Yayun Li, Xinran Ao, Qichao Chen, Zhengning Li, and Bin Meng. Road pothole extraction and safety evaluation by integration of point cloud and images derived from mobile mapping sensors.Advanced Engineering Informatics, 42:100936, 2019

  2. [10]

    Ride quality due to road surface irregularities: Comparison of different methods applied on a set of real road profiles.Coatings, 7(5):59, 2017

    Giuseppe Loprencipe and Pablo Zoccali. Ride quality due to road surface irregularities: Comparison of different methods applied on a set of real road profiles.Coatings, 7(5):59, 2017

  3. [11]

    Security issues in internet of vehicles (iov): A comprehensive survey.Internet of Things, 22:100809, 2023

    Hamideh Taslimasa, Sajjad Dadkhah, Euclides Carlos Pinto Neto, Pulei Xiong, Suprio Ray, and Ali A Ghorbani. Security issues in internet of vehicles (iov): A comprehensive survey.Internet of Things, 22:100809, 2023

  4. [12]

    Identification of problems faced in road maintenance.International Journal of Innovative Research in Engineering & Management, 10(3):29–37, 2023

    Sayim Niyaz Baba and Er Brahmjeet Singh. Identification of problems faced in road maintenance.International Journal of Innovative Research in Engineering & Management, 10(3):29–37, 2023

  5. [13]

    Performance analysis of image thresholding: Otsu technique.Measurement, 114:298– 307, 2018

    Ta Yang Goh, Shafriza Nisha Basah, Haniza Yazid, Muhammad Juhairi Aziz Safar, and Fathinul Syahir Ahmad Saad. Performance analysis of image thresholding: Otsu technique.Measurement, 114:298– 307, 2018

  6. [14]

    Pothole detection with image processing and spectral clustering

    Emir Buza, Samir Omanovic, and Alvin Huseinovic. Pothole detection with image processing and spectral clustering. InProceedings of the 2nd International Conference on Information Technology and Computer Networks, volume 810, page 4853, 2013

  7. [15]

    Pothole detection on asphalt pave- ments from 2d-colour pothole images using fuzzy c-means clustering and morphological reconstruction.Automation in Construction, 83:196– 211, 2017

    Yashon O Ouma and Michael Hahn. Pothole detection on asphalt pave- ments from 2d-colour pothole images using fuzzy c-means clustering and morphological reconstruction.Automation in Construction, 83:196– 211, 2017

  8. [16]

    Scale-adaptive pothole detection and tracking from 3-d road point clouds

    Rigen Wu, Jiahe Fan, Libo Guo, Lei Qiao, M Usman Maqbool Bhutta, Brett Hosking, Sergey Vityazev, and Rui Fan. Scale-adaptive pothole detection and tracking from 3-d road point clouds. In2021 IEEE International Conference on Imaging Systems and Techniques (IST), pages 1–5. IEEE, 2021

  9. [17]

    Road pothole detection based on crowdsourced data and extended mask r-cnn.IEEE Transactions on Intelligent Transportation Systems, 25(9):12504–12516, 2024

    Linchao Li, Jiazhen Liu, Jiabao Xing, Zhiyang Liu, Kai Lin, and Bowen Du. Road pothole detection based on crowdsourced data and extended mask r-cnn.IEEE Transactions on Intelligent Transportation Systems, 25(9):12504–12516, 2024

  10. [18]

    You only look once: Unified, real-time object detection

    Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object detection. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 779–788, 2016

  11. [19]

    Asphalt pavement pothole detection using deep learning method based on yolo neural network

    Ernin Niswatul Ukhwah, Eko Mulyanto Yuniarno, and Yoyon Kusnendar Suprapto. Asphalt pavement pothole detection using deep learning method based on yolo neural network. In2019 International Seminar on Intelligent Technology and Its Applications (ISITIA), pages 35–40. IEEE, 2019

  12. [20]

    Real-time pothole detection using deep learning.arXiv preprint arXiv:2107.06356, 2021

    Anas Al Shaghouri, Rami Alkhatib, and Samir Berjaoui. Real-time pothole detection using deep learning.arXiv preprint arXiv:2107.06356, 2021

  13. [21]

    Pothole detection and filling system using image processing and machine learning

    TC Mahalingesh, Harshit Mishra, RV Arun, Anshuman Anand, et al. Pothole detection and filling system using image processing and machine learning. In2024 International Conference on Smart Systems for applications in Electrical Sciences (ICSSES), pages 1–5. IEEE, 2024

  14. [22]

    Pothole mapping and patching quantity estimates using lidar-based mobile mapping systems

    Radhika Ravi, Ayman Habib, and Darcy Bullock. Pothole mapping and patching quantity estimates using lidar-based mobile mapping systems. Transportation Research Record, 2674(9):124–134, 2020

  15. [23]

    Siyuan Chen, Debra F Laefer, Xiangding Zeng, Linh Truong-Hong, and JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 15 Eleni Mangina. V olumetric pothole detection from uav-based imagery. Journal of Surveying Engineering, 150(2):05024001, 2024

  16. [24]

    Image-based pothole detection using multi-scale feature network and risk assessment.Electronics, 12(4):826, 2023

    Dong-Hoe Heo, Ji-Yoon Choi, Sang-Baeg Kim, Tae-Oh Tak, and Sheng- Peng Zhang. Image-based pothole detection using multi-scale feature network and risk assessment.Electronics, 12(4):826, 2023

  17. [25]

    Potholes detection using deep learning and area estimation using image processing

    Subash Kharel and Khaled R Ahmed. Potholes detection using deep learning and area estimation using image processing. InProceedings of SAI Intelligent Systems Conference, pages 373–388. Springer, 2021

  18. [26]

    Pothole detection and dimension estimation system using deep learning (yolo) and image processing

    Pranjal A Chitale, Kaustubh Y Kekre, Hrishikesh R Shenai, Ruhina Karani, and Jay P Gala. Pothole detection and dimension estimation system using deep learning (yolo) and image processing. In2020 35th International Conference on Image and Vision Computing New Zealand (IVCNZ), p...

  19. [27]

    On the integration of self-attention and convolution

    Xuran Pan, Chunjiang Ge, Rui Lu, Shiji Song, Guanfu Chen, Zeyi Huang, and Gao Huang. On the integration of self-attention and convolution. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 815–825, 2022

  20. [28]

    Bot-sort: Robust associations multi-pedestrian tracking.arXiv preprint arXiv:2206.14651, 2022

    Nir Aharon, Roy Orfaig, and Ben-Zion Bobrovsky. Bot-sort: Robust associations multi-pedestrian tracking.arXiv preprint arXiv:2206.14651, 2022

  21. [29]

    Pyramidal implementation of the affine lucas kanade feature tracker description of the algorithm.Intel corporation, 5(1-10):4, 2001

    Jean-Yves Bouguet et al. Pyramidal implementation of the affine lucas kanade feature tracker description of the algorithm.Intel corporation, 5(1-10):4, 2001

  22. [30]

    Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography.Communications of the ACM, 24(6):381–395, 1981

    Martin A Fischler and Robert C Bolles. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography.Communications of the ACM, 24(6):381–395, 1981

  23. [31]

    The hungarian method for the assignment problem

    Harold W Kuhn. The hungarian method for the assignment problem. Naval research logistics quarterly, 2(1-2):83–97, 1955

  24. [32]

    The use of lidar and artificial intelligence algorithms for detection and size estimation of potholes.Buildings, 14(4):1078, 2024

    Sk Abu Talha, Dmitry Manasreh, and Munir D Nazzal. The use of lidar and artificial intelligence algorithms for detection and size estimation of potholes.Buildings, 14(4):1078, 2024

  25. [33]

    Depth anything v2.arXiv preprint arXiv:2406.09414, 2024

    Lihe Yang, Bingyi Kang, Zilong Huang, Zhen Zhao, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything v2.arXiv preprint arXiv:2406.09414, 2024

  26. [34]

    Computer vision based pothole detection under challenging conditions

    Boris Bu ˇcko, Eva Lieskovsk´a, Katar´ına Z´abovsk´a, and Michal Z´abovsk`y. Computer vision based pothole detection under challenging conditions. Sensors, 22(22):8878, 2022

  27. [35]

    Yolov8 based deep learning method for potholes detection

    Shruti Kumari, Anjali Gautam, Suvramalya Basak, and Nidhi Saxena. Yolov8 based deep learning method for potholes detection. In2023 IEEE International Conference on Computer Vision and Machine Intelligence (CVMI), pages 1–6. IEEE, 2023

  28. [36]

    A real-time ghost machine learning model built on yolov8 for traffic road signs detection and classification in germany.Multimedia Systems, 30(6):344, 2024

    Mohammed Hussein and Wen-Xing Zhu. A real-time ghost machine learning model built on yolov8 for traffic road signs detection and classification in germany.Multimedia Systems, 30(6):344, 2024

  29. [37]

    Yolo v3-tiny: Object detection and recognition using one stage improved model

    Pranav Adarsh, Pratibha Rathi, and Manoj Kumar. Yolo v3-tiny: Object detection and recognition using one stage improved model. In2020 6th international conference on advanced computing and communication systems (ICACCS), pages 687–694. IEEE, 2020

  30. [38]

    An improved yolov3-spp algorithm for image-based pothole detection

    Tianxin Liu, Jiaxuan Li, Meiying Cai, Yuyong Cui, and Quan-Yong Fan. An improved yolov3-spp algorithm for image-based pothole detection. InInternational Symposium on Neural Networks, pages 328–

  31. [39]

    Real-time pothole detection using yolov5

    Sudhakar Ajmera, C Ashok Kumar, P Yakaiah, Bittu Kumar, and K Yashwanth Chowdary. Real-time pothole detection using yolov5. In 2022 International Conference on Advancements in Smart, Secure and Intelligent Computing (ASSIC), pages 1–5. IEEE, 2022

  32. [40]

    Yolov6: A single-stage object detection framework for industrial applications

    Chuyi Li, Lulu Li, Hongliang Jiang, Kaiheng Weng, Yifei Geng, Liang Li, Zaidan Ke, Qingyuan Li, Meng Cheng, Weiqiang Nie, et al. Yolov6: A single-stage object detection framework for industrial applications. arXiv preprint arXiv:2209.02976, 2022

  33. [41]

    A novel framework for pothole area estimation based on object detection and monocular metric depth estimation

    Dehao Wang, Yiwen Xu, Haohang Zhu, and Kaiqi Liu. A novel framework for pothole area estimation based on object detection and monocular metric depth estimation. In2024 IEEE International Con- ference on Signal, Information and Data Processing (ICSIDP), pages 1–6. IEEE, 2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.