Nighttime Vehicle Detection Algorithm Based on Improved YOLOv8

Nighttime vehicle detection is crucial for improving road traffic safety and driving experience. However, traditional detection systems face challenges in low-light conditions and are affected by vehicle lights, resulting in low detection accuracy. A night vehicle detection model based on YOLOv8 is proposed in this paper to solve this problem. We combine C2f module with an attention mechanism and introduce the upsampling operator CARAFE to enhance the fusion and exchange of feature information, thus improving the accuracy of nighttime vehicle detection. Additionally, the original regression loss function CIoU is improved to capture the shape characteristics of the bounding box more flexibly and enable the model to have stronger robustness and faster convergence speed. Experiments on nighttime traffic data sets demonstrate that the enhanced method retains real-time performance while significantly improving nighttime vehicle detection accuracy compared to the original YOLOv8 model.