Thermal Imaging-Based Vehicle Detection and Road Geometry Analysis System for Hairpin Bend Safety

Hill stations often witness road accidents, especially along curved stretches and limited visibility at hairpin bends increase the risk. Drivers experience a higher risk of collision due to vehicles approaching from blind spots, especially at night or in adverse weather conditions such as rain and fog, which are frequent in mountainous regions. Existing Internet of Things (IoT) solutions often have restricted coverage or diminished performance under these circumstances. This study introduces a system that combines thermal imaging and deep learning techniques for reliable vehicle detection and road geometry analysis. Thermal cameras provide dependable detection performance where traditional sensors fail to operate effectively. The YOLOv8 model achieves an overall mean average precision at 50% (mAP@50) of 85% and 92.3% for cars. A hybrid U-Net with a ResNet-18 encoder performs semantic segmentation to estimate road width and slope, showing a strong correlation with ground-truth measurements (Pearson r = 0.93). The system’s outputs enable adaptive real-time alerts via traffic lights, thereby enhancing safety at hairpin bends under low-visibility conditions.