Generative AI Strategies to Enhance Car Detection Under Adverse Weather Conditions

Computer vision is a pivotal technology of the current decade, enabling computers to interpret visual inputs, extract information, and make informed decisions. Core challenges in computer vision encompass detection, segmentation, and classification, with detection being crucial for applications like transportation systems. However, adverse weather conditions, such as rain, snow, and fog, pose significant challenges for object detection systems in real-world environments. Traditional approaches often incorporate denoising or enhancement modules to preprocess images. An alternative strategy involves training models using a mix of real and synthetic data. This method not only reduces data processing costs but also enhances model adaptability to various conditions. This paper focuses on enhancing the robustness and accuracy of car detection models under adverse weather conditions using the YOLO (You Only Look Once) object detection model. By leveraging generative AI to create synthetic weather condition data based on real images captured from city surveillance cameras, this paper developed a comprehensive dataset for training. Our contributions include a generative AI pipeline to simulate various weather scenarios and the integration of this synthetic data with real data to train the YOLO model. The results demonstrate improved performance and reliability of car detection systems in challenging environments, highlighting the efficacy of combining synthetic and real data for robust computer vision applications.

Keywords:

object detection , adverse weather , generative ai