Abstract: Recently, infrared object detection has played a crucial role in intelligent transportation and autonomous driving. Existing infrared object detection technologies face challenges in handling low-contrast scenes, detecting small objects, and ensuring real-time performance. KCL-YOLOv8s, an infrared road object detection algorithm based on the YOLOv8 framework, was proposed in this study. This algorithm introduced a module named C2f_LSK to replace the C2f module in the original algorithm, thereby enhancing the network's multiscale feature extraction capabilities. In addition, a feature-centric diffusion pyramid network (CDPN) was designed to enhance the ability of the model to detect small objects. Considering the limited computational resources available for vehicle-mounted systems, a lightweight shared convolutional head was designed to improve the detection efficiency of the model. Finally, an improved version of WIoU v3, termed WIoU v3*, was adopted as the loss function to replace the one in the original model, thereby improving the model's ability to classify and localize objects and enhancing its detection capability for small infrared targets. Experimental results showed that the improved model achieved a mean average precision (mAP0.5) of 70.3% on the FLIR_ADAS_v2 dataset, outperforming other state-of-the-art models. Compared with the YOLOv8s model, it achieved a 3.8% increase in mAP0.5, 26.2% reduction in the number of parameters, and 35.8% improvement in processing speed, making it more suitable for deployment in real-time, resource-constrained scenarios.