Adaptive Deep Learning Framework for Harsh Weather Traffic Sign Recognition

Abstract

Traffic sign recognition in adverse weather conditions poses a significant challenge to autonomous driving systems and demands robust and adaptive solutions. This project introduces a real-time traffic sign recognition framework designed to operate reliably under harsh weather scenarios by leveraging the adaptability and efficiency of deep learning techniques. optimized for embedded systems, for spatial feature extraction, and integrates a weather-adaptive preprocessing module to enhance image quality and maintain detection accuracy across varying environmental conditions. In advanced configurations, the framework incorporates attention mechanisms and domain adaptation strategies to dynamically adjust to visual distortions caused by rain, fog, snow, and low light settings. Upon successful recognition, the system updates contextual driving data including sign type, timestamp, and GPS coordinates, which are transmitted to the vehicle control unit and optionally to cloud services for analysis. All processed data are securely stored in cloud databases such as Firebase Fire store, enabling real-time monitoring and continuous model refinement.

Introduction

This project proposes an Adaptive Deep Learning Framework for Traffic Sign Recognition (TSR) under Harsh Weather Conditions to improve road safety in Advanced Driver-Assistance Systems (ADAS) and autonomous vehicles. While existing deep learning-based TSR systems perform well in clear weather, their accuracy decreases significantly in adverse conditions such as fog, rain, snow, blur, and low light due to reduced visibility, noise, and occlusions. To address these challenges, the framework incorporates adaptive techniques such as weather-specific preprocessing, data augmentation, domain adaptation, and robust feature learning.

The literature review highlights the evolution of traffic sign recognition from traditional machine learning approaches to CNN-based models, as well as advancements such as real-time detection, image enhancement, blurred image handling, fog-resistant recognition using CycleGAN, and feature enhancement through Squeeze-and-Excitation Networks.

The proposed methodology begins with real-time video capture from an onboard vehicle camera. Captured frames undergo preprocessing steps including resizing, normalization, and noise reduction. A weather classification module identifies environmental conditions such as fog, rain, snow, or clear weather. Based on the detected condition, adaptive enhancement techniques such as dehazing, brightness correction, and contrast enhancement are applied to improve image quality.

The enhanced images are then processed by a deep learning model, primarily MobileNetV2, with optional BiLSTM integration for sequential frame analysis. Traffic signs are detected and classified using a Softmax layer, generating labels and confidence scores. Post-processing removes noise and overlapping predictions before delivering the final results through on-screen displays and Text-to-Speech (TTS) alerts. Detection data can also be stored in a cloud database for analysis.

The system is implemented in phases, including real-time traffic sign detection, deep learning-based classification, harsh-weather recognition, image enhancement, and adaptive CNN-based prediction. Experimental results demonstrate that the proposed framework achieves higher accuracy, precision, recall, and F1-scores than traditional CNN models. It shows improved robustness, lower error rates, and better real-time performance in foggy, rainy, blurred, and low-light environments. Confusion matrix analysis further confirms more accurate classification and fewer misclassifications across traffic sign categories.

Conclusion

In conclusion, the proposed Adaptive Deep Learning Framework for Traffic Sign Recognition successfully improves the detection and classification of traffic signs under harsh weather conditions such as fog, rain, blur, and low illumination. By integrating deep learning techniques, image preprocessing, and weather-specific data augmentation, the system achieves higher accuracy, robustness, and real-time performance compared to traditional methods. The experimental results demonstrate the framework’s ability to maintain reliable recognition even in challenging environments, making it suitable for Advanced Driver-Assistance Systems (ADAS) and autonomous vehicles. Overall, this project contributes toward enhancing road safety and intelligent transportation systems through adaptive and efficient traffic sign recognition.

References

Huetal, [1] “Squeeze-and-Excitation Networks”: A Review. International Journal of Computer Applications, 181(28), 29-33. doi: 10.5120/ijca2019919551 Zhang[2] “Traffic Sign Recognition under Foggy Conditions Using Cycle AN”: International Journal of Advanced Research in Computer Science, 11(2), 1-5. Rattan, G [3] “Adaptive C`NNs for TSR in Adverse Weather” International Journal of Advanced Computer Science and Applications, 10(2), 103-110. Singh, D., & Rattan, G [4] “A Review of Traffic Sign Detection and Recognition Techniques. International Journal of Scientific Research in Computer Science”, Engineering and Information Technology, 4(2), 85-92. Temel, D., et al. “Challenging Environments for Traffic Sign Detection: Reliability Assessment under Inclement Conditions,” IEEE Signal Processing Society, 2019. Ahmed, S., Kamal, U., & Hasan, M. “DFR-TSD: A Deep Learning Based Framework for Robust Traffic Sign Detection Under Challenging Weather Conditions,” 2020. Zaki, P. S., et al. “Traffic Signs Detection and Recognition System using Deep Learning,” 2020.

Copyright

Copyright © 2026 Mrs. Nalini H C, Charanya K C, Rakshitha S S, Darshan M, Ajay C M. 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.