Enhancing Object Detection in Adverse Weather for Autonomous Driving with YOLOv9

The integration of autonomous vehicles (AVs) within the society has been a topic of interest since the 1950s when the trials on the first advanced driver assistance system (ADAS) began. The promise of autonomous driving systems is based on their ability to traverse with human-level perception, especially object detection in driving environments. This is essential for safe and reliable navigation but adverse weather conditions like sandstorms, rain, dense fog, and heavy snowfall hinder the robustness of the perception system. To address the problem of low visibility in adverse weathers like these, our proposed approach implemented feature extraction using content-based information retrieval (CBIR) for contrast enhancement and Enhanced Super Resolution Generative Adversarial Networks (ESRGAN) for image restoration. Further, YOLOv9 - the latest breakthrough in object detection was used. This work was the first initiative taken towards exploring the implementation of adverse weather driving dataset on Yolov9. The methodology was successful in achieving an average precision of 97.17% in detecting objects across all weathers with an overall mean Average Precision (mAP) of 72.87. These encouraging results might prove to be beneficial for safer and more reliable autonomous vehicle operation in diverse weather conditions.