<p>Infrared object detection is crucial for enhancing the safety of autonomous driving systems, especially under nighttime conditions. However, the existing infrared object detection models face challenges, including diminished detection efficiency, difficulties in achieving seamless integration with devices constrained by limited computational resources, and poor performance in detecting small objects in low-contrast infrared scenarios. To overcome these limitations, this study utilizes YOLOv8n as the baseline model and presents WBS-YOLOv8, a lightweight object detection network specifically designed to enhance performance and efficiency. The original loss function within the backbone network is substituted with a weighted intersection over union (WIoU) loss function, which incorporates a weighting mechanism to enhance the impact of medium-quality anchor boxes. A bi-directional feature pyramid network (BiFPN) is utilized to optimize the fusion of multi-scale features through bidirectional feature integration techniques. The spatial pyramid pooling fast with large spatial kernel attention (SPPF-LSKA) mechanism is introduced to further enhance detection accuracy and robustness in complex scenarios. Experimental results on the FLIR dataset reveal that the WBS-YOLOv8 model achieves a mAP@0.5 of 76.3%, and a mAP@[0.5:0.95] of 40.7% with merely 3.13 million parameters, representing increases of 1.7%, and 1.0%, respectively, over the baseline model. Moreover, WBS-YOLOv8 has also been successfully deployed on a low-power embedded platform, with a Frame Per Second (FPS) of 27. These findings collectively demonstrate that the WBS-YOLOv8 model can achieve real-time image processing in general scenes while maintaining high detection accuracy, particularly in challenging low-contrast and small object detection scenarios.</p>