Predictive Modeling of Vehicle Emissions Using Deep Learning Techniques and AERMOD

Real-time vehicle detection not only enables accurate counting and tracking of vehicles on the road but also provides valuable data for analyzing traffic patterns and vehicle-related emissions. Simultaneously, calculating vehicular emissions is crucial for assessing environmental impact and implementing pollution reduction strategies. These combined efforts play a pivotal role in enhancing transportation systems and creating a more sustainable infrastructure. In this study, we employed a real-time vehicle detection system based on deep learning. During rush hour traffic, vehicles traversing Beşiktaş-Barbaros Boulevard in Istanbul, Türkiye, were counted using the YOLO (You Only Look Once) version 8 (YOLOv8) and SORT (Simple Online and Real-time Tracking) algorithm based on video recordings. The system was trained using a custom ISTraffic dataset which includes 36,841 annotated instances across 1,125 images, covering various vehicle classes such as cars, shuttles, buses, and motorcycles. The proposed algorithm exhibited outstanding performance, achieving a mean average precision (mAP) of 0.903 at a confidence threshold of 0.5. It achieved a precision of 1.00 and a recall of 0.91. Furthermore, we calculated vehicle-related pollutant emissions, including CO, NO, NO₂,NO_x, and PM₁₀, using the COPERT program following the Tier 3 methodology. These emissions data were then used as input for the AERMOD model. AERMOD CO 2180 µg/m³, NO 2848 µg/m³,NO₂ 1373 µg/m³,NO_x 4222 µg/m³,PM₁₀ 155 µg/ m³ . By combining deep learning techniques with AERMOD, our research aims to improve predictive modeling of air quality impacts from transportation activities.