Deep Learning Approach for Detecting and Classifying Traffic Signs and Potholes in Indian Roads

Abstract

The Advanced Driver Assistance Systems (ADAS) have a significance in enhancing road safety by identifying the traffic signs and road surface problems in real-time. This research paper compares three models of object detection that are based on YOLO algorithm and are YOLOv5, YOLOv7, and YOLOv8 for detecting Indian traffic signs and potholes. The approach used in this research involved the steps of collecting dataset, annotating the data, processing the data, augmenting the data, and training the model. The results after conducting the experiments show that every model has its own particular functionalities. In case of traffic signs detection, YOLOv5 got the highest precision of 99.3 percent, however, YOLOv8 got the highest recall of 83.3 percent and mAP50 of 88.7 percent, showing the capability of identifying the sign on different road conditions. In case of pothole detection, YOLOv5 gave the best precision of 83.1 percent, but YOLOv7 performed best overall with retrieval of recall of 76.7 percent and mAP of 79.0 percent and mAP range from 50 to 95 of 48.3 percent. The comparison proves that the YOLO models of object detection are of great use in real-time ADAS and show good efficiency in road scene perception.

Introduction

This project focuses on developing an Advanced Driver Assistance System (ADAS) for Indian roads that detects traffic signs and potholes using You Only Look Once (YOLO) deep learning models. ADAS improves road safety by assisting drivers in identifying road hazards and reducing accidents caused by human error. The proposed system combines computer vision and machine learning to provide real-time detection of traffic signs and potholes, enhancing driving safety and comfort.

The literature survey reviews several traffic sign and pothole detection methods based on HOG, CNN, R-CNN, YOLOv2, YOLOv3, YOLOv5, YOLOv7, YOLOv8, ResNet50, LiDAR, and ultrasonic sensors. Previous studies reported high detection accuracies ranging from 84% to over 99%, but many were trained on non-Indian datasets or focused on either traffic signs or potholes separately. This highlights the need for a unified model trained on Indian road conditions.

The proposed methodology uses the YOLO object detection framework, which performs object localization and classification in a single stage, enabling fast and accurate real-time detection. The system consists of the following stages:

• Data Acquisition: Two datasets were collected—an Indian Cautionary Traffic Signs (ICTS) dataset containing 6,670 images and a pothole dataset containing 1,800 images. The datasets were divided into training, validation, and testing sets.
• Data Augmentation: Image transformations such as rotation and zooming were applied to increase dataset diversity and improve model performance.
• Model Development: The project evaluates three YOLO variants:
  • YOLOv5, which uses CSPDarknet, PANet, and multi-scale detection for high accuracy and speed.
  • YOLOv7, featuring ELAN/E-ELAN architectures for improved gradient flow, faster convergence, and enhanced detection performance.
  • YOLOv8, which incorporates a modified CSPDarknet backbone, self-attention mechanisms, and feature pyramid networks for improved multi-scale object detection.

Conclusion

The present work compared the performance of YOLOv5, YOLOv7, and YOLOv8 in terms of detecting potholes and traffic signs on Indian roads. The findings of the experiments demonstrated that although all three models produced good results, each of the models has some specific advantages. YOLOv5 achieved the highest precision; YOLOv7 peculiarly showed the best performance in terms of detecting potholes while YOLOv8 manifested high values of recall in detecting the traffic signs. Overall, it can be said that the YOLO models prove to be effective in the process of detecting hazards on the roads practically in real time or enhance the capabilities of ADAS. In the future, the performance of the system can be improved with the help of the incorporation of larger datasets and conducting the experiments in various weather conditions. Besides that, usage of larger datasets which consist of images taken in different lighting conditions can be selected in the project. Additionally, the system can be deployed on edge devices which can help making the process completely real-time.

References

[1] Merkle, P., Morvan, Y., Smolic, A., Farin, D., Mueller, K., De with, P. H. N., & Wiegand, T. (2009). The effects of multiview depth video compression on multiview rendering. Signal Processing: Image Communication, 24(1-2), 73-88. [2] Ge, S. S., & Wang, C. (2004). Adaptive neural control of uncertain MIMO nonlinear systems. IEEE Transactions on Neural Networks, 15(3), 674-692. [3] Wang, C. (2018, May). Research and application of traffic sign detection and recognition based on deep learning. In 2018 International Conference on Robots & Intelligent System (ICRIS) (pp. 150-152). IEEE. [4] Fernando, W. H. D., & Sotheeswaran, S. (2021, September). Automatic road traffic signs detection and recognition using ‘You Only Look Once’version 4 (YOLOv4). In 2021 International Research Conference on Smart Computing and Systems Engineering (SCSE) (Vol. 4, pp. 38-43). IEEE. [5] Zhang, Z., Xin, L., Sun, K., & Li, W. (2011). Pd–Ni electrocatalysts for efficient ethanol oxidation reaction in alkaline electrolyte. International Journal of Hydrogen Energy, 36(20), 12686-12697. [6] Sharma, A., Bhatt, N. S., St Martin, A., Abid, M. B., Bloomquist, J., Chemaly, R. F., ... & Shah, G. L. (2021). Clinical characteristics and outcomes of COVID-19 in haematopoietic stem- cell transplantation recipients: an observational cohort study. The Lancet Haematology, 8(3), e185-e193. [7] Oza, R. M., Geisen, A., & Wang, T. (2021, September). Traffic Sign Detection and Recognition using Deep Learning. In 2021 4th International Conference on Artificial Intelligence for Industries (AI4I) (pp. 16-20). IEEE. [8] Zarrouk, A., Vejux, A., Mackrill, J., O’Callaghan, Y., Hammami, M., O’Brien, N., & Lizard, G. (2014). Involvement of oxysterols in age-related diseases and ageing processes. Ageing research reviews, 18, 148-162. [9] Tabernik, D., & Sko?aj, D. (2019). Deep learning for large-scale traffic-sign detection and recognition. IEEE transactions on intelligent transportation systems, 21(4), 1427-1440. [10] Dhar, P., Abedin, M. Z., Biswas, T., & Datta, A. (2017, December). Traffic sign detection— A new approach and recognition using convolution neural network. In 2017 IEEE Region 10 Humanitarian Technology Conference (R10-HTC) (pp. 416-419). IEEE. [11] Reddy, E. S. T. K., & Rajaram, V. (2022, December). Pothole Detection using CNN and YOLO v7 Algorithm. In 2022 6th International Conference on Electronics, Communication and Aerospace Technology (pp. 1255-1260). IEEE. [8] Madli, R., Hebbar, S., Pattar, P., & Golla, V. (2015). Automatic detection and notification of potholes and humps on roads to aid drivers. IEEE sensors journal, 15(8), 4313-4318. [12] Saranya, G., & Pravin, A. (2020). A comprehensive study on disease risk predictions in machine learning. International Journal of Electrical and Computer Engineering, 10(4), 4217. [13] Kang, B. H., & Choi, S. I. (2017, July). Pothole detection system using 2D LiDAR and camera. In 2017 ninth international conference on ubiquitous and future networks (ICUFN) (pp. 744-746). IEEE. [14] Singh, S., & Gupta, P. (2014). Comparative study ID3, cart and C4. 5 decision tree algorithm: a survey. International Journal of Advanced Information Science and Technology (IJAIST), 27(27), 97-103. [15] Dharneeshkar, J., Aniruthan, S. A., Karthika, R., & Parameswaran, L. (2020, February). Deep Learning based Detection of potholes in Indian roads using YOLO. In 2020 international conference on inventive computation technologies (ICICT) (pp. 381-385). IEEE. [16] Kuan, C. W., Chen, W. H., & Lin, Y. C. (2020, November). Pothole detection and avoidance via deep learning on edge devices. In 2020 International Automatic Control Conference (CACS) (pp. 1-6). IEEE. [17] Rohitaa, R., Shreya, S., & Amutha, R. (2021, September). Intelligent Deep Learning based Pothole Detection and Reporting System. In 2021 Fourth International Conference on Electrical, Computer and Communication Technologies (ICECCT) (pp. 1-5). IEEE [18] Madli, R., Hebbar, S., Pattar, P., & Golla, V. (2015). Automatic detection and notification of potholes and humps on roads to aid drivers. IEEE sensors journal, 15(8), 4313-4318. [19] Wang, X., Hegde, S., Son, C., Keller, B., Smith, A., & Sasangohar, F. (2020). Investigating mental health of US college students during the COVID-19 pandemic: Cross-sectional survey study. Journal of medical Internet research, 22(9), e22817. [20] Meng, Y., Xu, J., Jin, Z., Prakash, B., & Hu, Y. (2020). A review of recent advances in tribology. Friction, 8, 221-300.

Copyright

Copyright © 2026 Gandu Gayathri, Ch Chandrika. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.