Enhancing Robust Object Detection in Weather-Impacted Environments using Deep Learning Techniques

Abstract

This research study presents R-YOLO (Robust You Only Look Once), a robust object detection framework designed to address weather-related issues. R-YOLO makes use of the YOLO architecture and incorporates additional features to lessen noise and increase visibility in the event of adverse weather. Adaptive filtering and a weather-aware attention mechanism are two of these features. By combining conventional datasets with a well annotated weather-specific dataset, the model is trained on a sizable dataset that spans a variety of weather conditions. According to the trial results, R-YOLO outperforms both baseline YOLO models and state-of-the-art detectors under adverse weather conditions, exhibiting more adaptability and generalizability. Furthermore, real-time applications like robotics, autonomous driving, and surveillance are suitable for R-YOLO. The proposed framework not only maintains good performance in a variety of adverse weather conditions but also improves object detection accuracy in low visibility scenarios, making it a dependable and useful choice for real-world implementation. This paper investigates the potential benefits of R- YOLO for computer vision applications, particularly in situations where adverse weather poses significant challenges.