Potholes and unmaintained pavements lead to dis- rupted transportation that becomes unsafe and more costly to maintain. Repairing damage takes these prices up, delays traffic, and brings about a higher accident risk for the so-called most vulnerable road users, such as cyclists and motorcycle riders. They also affect travel and logistics, thus the dire need for monitoring and timely remedies. The usual manual techniques suchasthe surfacedistressindexhavebeenused inroaddamage assessment.However,itislimitedwithrespecttohumanendeavor andtimeandscalability.Theyincludeautomatedtechniques for analyzing the road surface. Examples of successful methods adopted are the K-Nearest Neighbours (KNN)[21] coupled with theGrayLevelCo-occurrenceMatrix(GLCM)[20]todetect and classify road defects. Likewise, Support Vector Machines (SVMs) have been used effectively in classifying potholes, using partial differential equations for image segmentation. However, although these methods may work well, they have limitations, such as being very dependent on the datasets and being sensitive to the lighting conditions during capture. Recent advances in deeplearning,suchasthefamilyofmodelsofYouOnly Look Once (YOLO)[12], have provided good opportunities to improve these pothole detection systems concerning speed and accuracy. This work aims to continue by improving the processofpotholedetectionandsegmentationforgreaterefficiency,wider applicability to the detection of potholes, and more accurate damage assessments. This ultimately facilitates effective repairs and maintenance processes.
Introduction
Overview:
Modern research into pothole detection has made extensive use of machine learning (ML) and deep learning (DL) models, with a focus on cost-effective, real-time systems for road maintenance and safety improvement. The literature explores different models, including CNNs, YOLO variants, Faster R-CNN, and hybrid approaches combining sensors and image processing.
Key Methods and Technologies:
Smartphone-Based Detection:
Utilizes gyroscope and accelerometer data processed through InceptionV3 with transfer learning.
Offers an efficient, scalable, and low-cost solution.
Image-Based Detection Using DL Models:
Faster R-CNN + InceptionV2: Uses region proposals for image/video-based pothole detection.
YOLOv8n-seg: Lightweight, highly accurate object detection and segmentation model; supports advanced augmentations for robust detection.
KNN + GLCM: Effective in classifying surface defects using pixel-level texture analysis.
Recent Advances:
YOLOv9 & YOLOv11: Introduced for improved accuracy using features like Programmable Gradient Information (PGI) and GELAN for better training of lightweight models.
Major Contributions of the Study:
Addresses accuracy limitations of prior models (e.g., KNN-GLCM).
Enhances detection under challenging conditions (e.g., lighting, occlusion).
Focuses on data augmentation to improve model generalization.
Proposes real-world deployment strategy with action plans for maintenance response.
Integrates performance-driven optimization techniques to boost segmentation/classification.
Literature Review Highlights:
Thantharate et al. (2024): Compared YOLOv5–v8 and RT-DETR, showing YOLOv8 as best for real-time; RT-DETR offers higher accuracy at higher computation cost.
Jasmine Hephzipah et al. (2023): Developed a sensor-robotic system with GPS and GSM for road quality monitoring.
Munish Rathee et al. (2023): Explored ARDAD frameworks combining sensor fusion with AI for accident prevention.
Senthil Kumar et al. (2023): Used ML on acoustic data with up to 98.98% accuracy; relays road conditions in real-time.
Zhou et al. (2022): Reviewed sensor and imaging technologies, emphasized the need for better data platforms and rural road consideration.
Lubis et al. (2022): KNN-GLCM image analysis improved damage classification and real-time infrastructure management.
Ahmed (2021): Proposed MVGG16 + Faster R-CNN; balanced accuracy and computation time.
Kumar et al. (2020): Used Inception-V2 + Faster R-CNN; high accuracy (99.8%) and low training time.
Al-Shaghouri et al. (2020): YOLOv4 (mAP: 85.39%) enabled real-time detection and potential integration into autonomous systems.
Dhiman & Klette (2019): Combined stereo vision and DL for 3D pothole detection; emphasized dataset standardization.
Song et al. (2018): Smartphone solution using InceptionV3; achieved 100% accuracy under tested conditions.
Proposed Framework in This Study:
Problem Statement: Existing models often struggle with false positives/negatives under complex conditions.
Solution: Advanced deep learning (YOLOv9 & YOLOv11) with real-time processing and robust dataset handling.
Workflow:
Data Collection: Images from varied conditions (lighting, surface types).
Annotation: Bounding boxes for potholes and severity metrics.
Model Training: Utilizing YOLOv9 and YOLOv11 for better detection and severity classification.
YOLOv9 Features:
Programmable Gradient Information (PGI): Minimizes information loss during training.
GELAN (Generalized Efficient Layer Aggregation Network): Enhances lightweight model performance.
Conclusion
Thisreviewwillfocusonthelatestadvancementsinpothole detection and damage assessment. It will discuss deep learn- ing models including YOLOv9 and YOLOv11 that have the potentialtoreplacecurrentmethodologies.Ourworkanalyzes the existing methodologies and highlights their disadvantages to develop a robust and accurate real-time pothole detection and severity evaluation model. The introduced machines will further enhance accuracy, speed, and adaptability with respect tovariousroadconditionsbyemployinginnovativefeatures.
The performance in terms of application scenarios has been validated for YOLOv9 and YOLOv11 by comparing their performance through various real-field contexts where the models can be demonstrated with their own strengths and weaknesses. The research objectives, apart from evaluatingthe accuracy and robustness of the models, will also lay the foundation for future work in developing hybrid models that could combine some of the best attributes of both worlds. Our workwillthereforecontributetosafety,facilitatemaintenance workflows in infrastructure, and address the broad challenge areas in transportation management in scalable and efficient ways.
References
[1] Maglogiannis, L. Iliadis, J. Macintyre, M. Avlonitis, and A. Papale-onidas, ”Artificial Intelligence Applications and Innovations: 20th IFIPWG 12.5 International Conference, AIAI 2024, Corfu, Greece, June27–30, 2024, Proceedings, Part III,” IFIP Advances in Information andCommunication Technology, vol. 713, pp. 1–16, Springer, 2024.
[2] J. J. Hephzipah, B. Sarala, M. Perarasi, K. S. Kowsik, M. P. Mano- jKummar, and S. Kogulram, ”Road Crack and Road Quality Checking Mechanism,” Proc. 7th Int. Conf. Intell. Comput. Control Syst., IEEE,
[3] pp. 882–886, 2023.
[4] M. Rathee, B. Bac?ic´, and M. Doborjeh, ”Automated Road Defect and Anomaly Detection for Traffic Safety: A Systematic Review,” Sensors, vol. 23, no. 12, p. 5656, June 2023.
[5] S. K. Jagatheesaperumal, S. E. Bibri, S. Ganesan, and P. Jeyaraman, ”Artificial Intelligence for Road Quality Assessment in Smart Cities: A Machine Learning Approach to Acoustic Data Analysis,” Computational Urban Science, vol. 3, no. 28, Art. ID 104, 2023.
[6] Y. Zhou, X. Guo, F. Hou, and J. Wu, ”Review of Intelligent Road Defects Detection Technology,” Sustainability, vol. 14, no. 6306, Art. ID 6306, 2022.
[7] A. Lubis, I. Iskandar, and M. L. W. Panjaitan, ”Implementation of KNN Methods and GLCM Extraction for Classification of Road Damage Level,” IAIC Transactions on Sustainable Digital Innovation (ITSDI), vol. 4, no. 1, Art. ID 564, 2022.
[8] K. R. Ahmed, ”Smart Pothole Detection Using Deep Learning Based on Dilated Convolution,” Sensors, vol. 21, no. 24, Art. ID 8406, Dec. 2021.
[9] A. Kumar, V. P. Singh, Chakrapani, and D. J. Kalita, ”A Modern Pothole Detection Technique Using Deep Learning,” in Proceedings of IEEE, Art. ID August 26, 2020.
[10] A. Al-Shaghouri, R. Alkhatib, and S. Berjaoui, ”Real-Time Pothole De- tection Using Deep Learning,” Mathematical Problems in Engineering, vol. 2020, Art. ID 4052672, 2020.
[11] A. Dhiman and R. Klette, ”Pothole Detection Using Computer Vision and Learning,” IEEE Transactions on Intelligent Transportation Systems, vol. XX, no. XX, pp. 1–14, IEEE, 2019.
[12] H. Song, K. Baek, and Y. Byun, ”Pothole Detection using Machine Learning,” Advanced Science and Technology Letters, vol. 150, pp. 151–155, SERSC, 2018.
[13] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, ”You Only Look Once: Unified, Real- Time Object Detection,” arXiv preprint arXiv:1506.02640, 2016.
[14] C.-Y. Wang, I.-H. Yeh, and H.-Y. M. Liao, ”YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information,” arXiv preprint arXiv:2402.13616v2, 2024.
[15] R. Khanam and M. Hussain, ”YOLOv11: An Overview of the Key Architectural Enhancements,” arXiv preprint arXiv:2410.17725v1, 2024.
[16] WenyuLv, ShangliangXu, Yian Zhao, Guanzhong Wang, Jinman Wei, Cheng Cui, Yuning Du, Qingqing Dang, and Yi Liu. DETRs beat YOLOs on real-time object detection. arXiv preprint arXiv:2304.08069, 2023.
[17] Yuming Chen, Xinbin Yuan, Ruiqi Wu, Jiabao Wang, QibinHou, and Ming-Ming Cheng. YOLO-MS: rethinking multi scale representation learning for real-time object detection. arXiv preprint arXiv:2308.05480, 2023.
[18] Ultralytics. Ultralytics yolov11. https://docs.ultralytics.com/models/yolo11/s, 2024. Accessed: 21-Oct-2024.
[19] D. Reis, J. Hong, J. Kupec, and A. Daoudi, ”Real-Time Flying Object Detection with YOLOv8,” arXiv preprint arXiv:2305.09972v2, 2024.
[20] N. Chandra, H. Vaidya, S. Sawant, and S.R. Meena, ”A Novel Attention- Based Generalized Efficient Layer Aggregation Network for Landslide Detection from Satellite Data in the Higher Himalayas, Nepal,” Remote Sensing, vol. 16, no. 2598, pp. 1–19, 2024.
[21] B. Sebastian V., A. Unnikrishnan, and K. Balakrishnan, ”Grey Level Co-occurrence Matrices: Generalisation and Some New Features,” In- ternational Journal of Computer Science, Engineering and Information Technology (IJCSEIT), vol. 2, no. 2, pp. 151–157, April 2012.
[22] G. Guo, H. Wang, D. A. Bell, and Y. Bi, ”KNN Model-Based Approach in Classification,” ODBASE Proceedings, August 2004.