pith. sign in

arxiv: 2508.10945 · v2 · submitted 2025-08-13 · 💻 cs.CV · cs.LG

iWatchRoad: Scalable Detection and Geospatial Visualization of Potholes for Smart Cities

Rishi Raj Sahoo , Surbhi Saswati Mohanty , Subhankar Mishra This is my paper

Pith reviewed 2026-05-18 22:44 UTC · model grok-4.3

classification 💻 cs.CV cs.LG
keywords pothole detectionYOLOgeospatial visualizationdashcamOpenStreetMaproad maintenancesmart citiescomputer vision
0
0 comments X

The pith

iWatchRoad combines a fine-tuned YOLO detector with GPS and OCR to turn ordinary dashcam video into geotagged pothole maps on OpenStreetMap.

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

The paper presents iWatchRoad as an end-to-end pipeline that detects potholes from vehicle videos, records their exact locations, and displays them on public maps. The authors created a dataset of more than 7,000 frames from many Indian road types, lighting levels, and weather conditions, then used it to adapt the Ultralytics YOLO model for real-time detection. An OCR module reads the video clock and matches it to separate GPS logs so every detection receives accurate latitude, longitude, and time. The results, including the original frames, are stored and shown through a web interface built on OpenStreetMap, producing records that road authorities can use for maintenance planning. If this works as described, cities gain a low-cost way to monitor road damage without special equipment or manual inspection.

Core claim

iWatchRoad demonstrates that fine-tuning Ultralytics YOLO on a self-annotated dataset of over 7,000 frames captured across varied Indian road types, lighting conditions, and weather scenarios, together with OCR timestamp extraction and GPS synchronization, produces accurate real-time pothole detections that can be stored with metadata and visualized on OpenStreetMap to support government road assessment and maintenance planning.

What carries the argument

The iWatchRoad pipeline that links fine-tuned YOLO object detection, OCR-based timestamp reading from video frames, GPS log synchronization for geotagging, database storage of detections and frames, and OpenStreetMap web visualization.

If this is right

  • Maintenance teams receive location-tagged images and timestamps that can be used directly for repair scheduling.
  • The same hardware and software setup works for both urban streets and rural roads without extra sensors.
  • New footage can be processed continuously to update the map as vehicles drive.
  • The outputs match the format needed for official road condition reports in developing regions.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The pipeline could be retrained to spot additional road defects such as cracks or loose gravel.
  • Multiple vehicles running the same software would create crowdsourced coverage of entire cities or highways.
  • Historical detections over months or years might reveal which road sections deteriorate fastest.
  • Deployment on roads outside India would require checking whether the current training data still suffices.

Load-bearing premise

The self-annotated dataset of over 7,000 frames from varied Indian road types, lighting, and weather is representative enough for the fine-tuned YOLO model to generalize reliably in real-world deployment.

What would settle it

A field test on dashcam footage from road conditions absent from the training set, such as heavy monsoon flooding or completely unpaved surfaces, that shows detection accuracy falling well below the levels reported in the paper.

Figures

Figures reproduced from arXiv: 2508.10945 by Rishi Raj Sahoo, Subhankar Mishra, Surbhi Saswati Mohanty.

Figure 1
Figure 1. Figure 1: This simplified pipeline showcases the core stages of the [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: A comprehensive dataset covering diverse road types, weather, and lighting conditions to fine-tune the model for improved [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison between traditional top-down pothole dataset (a), and the proposed forward facing dashcam view (b), which [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Shadow misclassification as a pothole demonstrates the necessity of comprehensive negative sample training for robust [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Model’s performance on BharatPotHole-3K [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Model’s performance on BharatPotHole-7K 4.2 Detection The detection setup comprises three key modules: 4.2.1 Custom trained YOLOv8 Detector: The model is fine-tuned using the BharatPotHole dataset, which includes thousands of annotated frames reflecting real time Indian road conditions, such as poorly lit, rain affected, or unpaved roads. The model’s performance increased drastically with an increase in th… view at source ↗
Figure 7
Figure 7. Figure 7: iWatchRoad’s Web Platform Interface: Interactive map view showing geotagged pothole reports with color coding based on severity,(1) shows a cluster of potholes, and when clicked, it shows (2) detailed metadata for each pothole, including timestamp, location, and frame image. (3) Upload interface where users can submit dashcam video and GPS files for automatic pothole detection and mapping. (4) Enumerates t… view at source ↗
Figure 8
Figure 8. Figure 8: The diagram illustrates the end-to-end workflow of our pothole detection system. Videos captured via dashcams are converted [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
read the original abstract

Potholes on the roads are a serious hazard and maintenance burden. This poses a significant threat to road safety and vehicle longevity, especially on the diverse and under-maintained roads of India. In this paper, we present a complete end-to-end system called iWatchRoad for automated pothole detection, Global Positioning System (GPS) tagging, and real time mapping using OpenStreetMap (OSM). We curated a large, self-annotated dataset of over 7,000 frames captured across various road types, lighting conditions, and weather scenarios unique to Indian environments, leveraging dashcam footage. This dataset is used to fine-tune, Ultralytics You Only Look Once (YOLO) model to perform real time pothole detection, while a custom Optical Character Recognition (OCR) module was employed to extract timestamps directly from video frames. The timestamps are synchronized with GPS logs to geotag each detected potholes accurately. The processed data includes the potholes' details and frames as metadata is stored in a database and visualized via a user friendly web interface using OSM. iWatchRoad not only improves detection accuracy under challenging conditions but also provides government compatible outputs for road assessment and maintenance planning through the metadata visible on the website. Our solution is cost effective, hardware efficient, and scalable, offering a practical tool for urban and rural road management in developing regions, making the system automated. iWatchRoad is available at https://smlab.niser.ac.in/project/iwatchroad

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript presents iWatchRoad, an end-to-end system for automated pothole detection on Indian roads. It curates a self-annotated dataset of over 7,000 dashcam frames across varied road types, lighting, and weather; fine-tunes an Ultralytics YOLO model for real-time detection; uses a custom OCR module to extract timestamps from video frames; synchronizes these with GPS logs for accurate geotagging; stores results with metadata in a database; and visualizes them on OpenStreetMap via a user-friendly web interface. The work claims improved detection accuracy under challenging conditions and provides government-compatible outputs for road assessment and maintenance planning, positioning the system as cost-effective, hardware-efficient, and scalable for smart cities in developing regions.

Significance. If the accuracy claims are substantiated, the integration of real-time detection with precise geotagging and public geospatial visualization offers a practical, deployable tool for road maintenance in under-resourced areas. The emphasis on metadata suitable for government use and the availability of the system at a project website are positive for real-world applicability and reproducibility. The approach assembles existing components (YOLO, OCR, GPS, OSM) in a domain-specific pipeline rather than introducing novel algorithms.

major comments (2)
  1. [Abstract] Abstract: The central claim that iWatchRoad 'improves detection accuracy under challenging conditions' is unsupported by any quantitative results. No mAP, precision, recall, F1-score, confusion matrix, train/test split details, baseline comparisons (e.g., to off-the-shelf YOLO or prior pothole detectors), or ablation studies on the 7,000-frame dataset are provided. This absence prevents evaluation of the detection step, which is load-bearing for all downstream claims about the pipeline and its utility.
  2. [Dataset and Model Fine-tuning] Dataset curation and model fine-tuning description: The self-annotated dataset of over 7,000 frames is presented as representative of Indian road conditions, but no information is given on annotation protocol, quality control, inter-annotator agreement, class distribution, or how 'challenging conditions' (lighting, weather, road types) were balanced or tested. This directly affects the weakest assumption regarding reliable generalization in real-world deployment.
minor comments (2)
  1. The project website link is provided but the manuscript does not indicate whether the dataset, code, or trained model weights are publicly released to support reproducibility.
  2. Figure captions and the web-interface description could more explicitly link displayed metadata fields to the government-compatible outputs claimed in the abstract.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment below and will revise the manuscript to provide the requested quantitative details and methodological clarifications.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that iWatchRoad 'improves detection accuracy under challenging conditions' is unsupported by any quantitative results. No mAP, precision, recall, F1-score, confusion matrix, train/test split details, baseline comparisons (e.g., to off-the-shelf YOLO or prior pothole detectors), or ablation studies on the 7,000-frame dataset are provided. This absence prevents evaluation of the detection step, which is load-bearing for all downstream claims about the pipeline and its utility.

    Authors: We agree that the abstract and current manuscript do not include quantitative performance metrics or comparisons to support the claim of improved detection accuracy. The manuscript emphasizes the end-to-end pipeline and deployment aspects but lacks a dedicated evaluation section. We will revise the abstract to incorporate key metrics and add a new results section reporting mAP, precision, recall, F1-score, confusion matrix, train/test split details (e.g., 80/20), baseline comparisons against the off-the-shelf Ultralytics YOLO and relevant prior pothole detectors, plus ablation studies on the effects of lighting, weather, and road types. revision: yes

  2. Referee: [Dataset and Model Fine-tuning] Dataset curation and model fine-tuning description: The self-annotated dataset of over 7,000 frames is presented as representative of Indian road conditions, but no information is given on annotation protocol, quality control, inter-annotator agreement, class distribution, or how 'challenging conditions' (lighting, weather, road types) were balanced or tested. This directly affects the weakest assumption regarding reliable generalization in real-world deployment.

    Authors: We acknowledge the absence of these dataset details in the manuscript. We will expand the dataset and model fine-tuning section to describe the annotation protocol (multi-annotator use of tools such as LabelImg), quality control steps, inter-annotator agreement statistics, class distribution across the 7,000 frames, and the sampling approach used to balance representation of diverse lighting, weather, and road-type conditions. revision: yes

Circularity Check

0 steps flagged

No significant circularity; engineering assembly of standard components with no derivations or self-referential claims

full rationale

The paper presents an applied system for pothole detection and mapping: it curates a self-annotated dataset of 7000+ frames, fine-tunes an off-the-shelf Ultralytics YOLO model, extracts timestamps via custom OCR, synchronizes with GPS logs, and visualizes results on OSM. No equations, first-principles derivations, or predictions appear in the abstract or described pipeline. The accuracy improvement claim is an unverified assertion rather than a derived result, but it does not reduce to any input by construction, self-citation load-bearing, or renamed known pattern. The work is self-contained as a practical integration of existing tools without any load-bearing step that collapses to its own fitted values or prior author results.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Applied computer-vision system paper; contains no mathematical derivations, free parameters, axioms, or invented entities beyond standard use of pre-existing tools (YOLO, OCR, OSM, GPS).

pith-pipeline@v0.9.0 · 5812 in / 1064 out tokens · 36021 ms · 2026-05-18T22:44:34.528422+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages

  1. [1]

    Jaided AI. 2020. EasyOCR: Ready-to-use OCR with 80+ languages supported. https://github.com/JaidedAI/EasyOCR

    work page 2020

  2. [2]

    Kranthi Kumar Chowdary, Y

    S. Kranthi Kumar Chowdary, Y. Harshith, and T. Preethiya. 2025. Smart Pothole Detection and Traffic Sign Identification for Indian Roads: A Machine Learning Approach Using Yolov11. In 2025 International Conference on Data Science and Business Systems (ICDSBS) . 1–6. doi:10.1109/ ICDSBS63635.2025.11031494

    work page arXiv 2025

  3. [3]

    Brad Dwyer, Joseph Nelson, Tom Hansen, et al. 2025. Roboflow (Version 1.0). https://roboflow.com Computer vision platform

    work page 2025

  4. [4]

    Rui Fan, Yanan Liu, Xingrui Yang, Mohammud Junaid Bocus, Naim Dahnoun, and Scott Tancock. 2018. Real-Time Stereo Vision for Road Surface 3-D Reconstruction. In 2018 IEEE International Conference on Imaging Systems and Techniques (IST) . 1–6. doi:10.1109/IST.2018.8577119

    work page doi:10.1109/ist.2018.8577119 2018

  5. [5]

    Hoseini, S

    M. Hoseini, S. Puliti, S. Hoffmann, and R. Astrup. 2023. Pothole detection in the woods: a deep learning approach for forest road surface monitoring with dashcams. International Journal of Forest Engineering 35, 2 (2023), 303–312. doi:10.1080/14942119.2023.2290795

    work page doi:10.1080/14942119.2023.2290795 2023

  6. [6]

    Dharneeshkar J, Soban Dhakshana V, Aniruthan S A, Karthika R, and Latha Parameswaran. 2020. Deep Learning based Detection of potholes in Indian roads using YOLO. In 2020 International Conference on Inventive Computation Technologies (ICICT) . 381–385. doi:10.1109/ICICT48043.2020.9112424

    work page doi:10.1109/icict48043.2020.9112424 2020

  7. [7]

    Glenn Jocher, Ayush Chaurasia, and Jing Qiu. 2023. Ultralytics YOLOv8. https://github.com/ultralytics/ultralytics

    work page 2023

  8. [8]

    Lincy, G

    A. Lincy, G. Dhanarajan, S. Sanjay Kumar, and B. Gobinath. 2023. Road Pothole Detection System. ITM Web Conf. 53 (2023), 01008. doi:10.1051/ itmconf/20235301008

    work page arXiv 2023

  9. [9]

    Nachuan Ma, Jiahe Fan, Wenshuo Wang, Jin Wu, Yu Jiang, Lihua Xie, and Rui Fan. 2022. Computer vision for road imaging and pothole detection: a state-of-the-art review of systems and algorithms. Transportation Safety and Environment 4, 4 (Nov. 2022). doi:10.1093/tse/tdac026

    work page doi:10.1093/tse/tdac026 2022

  10. [10]

    Mohd Omar and Pradeep Kumar. 2024. PD-ITS: Pothole Detection Using YOLO Variants for Intelligent Transport System. SN Comput. Sci. 5, 5 (May 2024), 16 pages. doi:10.1007/s42979-024-02887-1

    work page doi:10.1007/s42979-024-02887-1 2024

  11. [11]

    OpenStreetMap contributors. 2024. Planet dump retrieved from https://planet.openstreetmap.org. https://www.openstreetmap.org

    work page 2024

  12. [12]

    Alfandino Rasyid, Mochammad Rifki Ulil Albaab, Muhammad Fajrul Falah, Yohanes Yohanie Fridelin Panduman, Alviansyah Arman Yusuf, Dwi Kurnia Basuki, Anang Tjahjono, Rizqi Putri Nourma Budiarti, Sritrusta Sukaridhoto, Firman Yudianto, and Hendro Wicaksono. 2019. Pothole Visual Detection using Machine Learning Method integrated with Internet of Thing Video S...

    work page doi:10.1109/elecsym.2019.8901626 2019

  13. [13]

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

    work page 2016

  14. [14]

    Yeganeh Safyari, Masoud Mahdianpari, and Hamid Shiri. 2024. A Review of Vision-Based Pothole Detection Methods Using Computer Vision and Machine Learning. Sensors 24, 17 (2024), 5652. doi:10.3390/s24175652

    work page doi:10.3390/s24175652 2024

  15. [15]

    Amxson Sminage, Delvin P B, Derick Davies, Vivek K J, and Jasmy Davies. 2025. SafeDrive: Intelligent Pothole Detection and Mapping System. In 2025 2nd International Conference on Trends in Engineering Systems and Technologies (ICTEST) , Vol. 1. 1–6. doi:10.1109/ICTEST64710.2025.11042541

    work page doi:10.1109/ictest64710.2025.11042541 2025

  16. [16]

    Javier Yebes, David Montero, and Ignacio Arriola

    J. Javier Yebes, David Montero, and Ignacio Arriola. 2021. Learning to Automatically Catch Potholes in Worldwide Road Scene Images. IEEE Intelligent Transportation Systems Magazine 13, 3 (2021), 192–205. doi:10.1109/mits.2019.2926370

    work page doi:10.1109/mits.2019.2926370 2021