In the context of infrared vehicle and pedestrian recognition applications, the detection algorithm may encounter challenges due to image blurring and scene diversity. This can result in issues such as occlusion and the misdetection or misidentification of distant objects. This paper puts forth an enhanced YOLOv8 vehicle and pedestrian object detection algorithm, designated as STF-YOLO. Firstly, to address the difficulty of extracting target features from infrared low-resolution images, two new techniques, SPDConv (Space to Depth Convolution) and TA (Triple Attention), are introduced into the model backbone with the objective of enhancing the model’s ability to recognize fine features. Secondly, the method of fusing the neck network has been enhanced. This has led to an improvement in the multi-scale fusion ability of the model, as well as an enhancement in the ability to extract subtle feature information from low-level feature maps. Finally, the characteristics of the Focal-GIOU loss function have been employed to eliminate the imbalance between the high- and low-quality sample categories. This has enhanced the positional precision of the prediction frame and a promotion of the speed of algorithm convergence. The mAP (mean Average Precision) of STF-YOLO on the FLIR dataset was found to be 80.3%, representing a 4.1% improvement over the benchmark algorithm, YOLOv8n. Furthermore, the algorithm’s inference speed is sufficient for the timely completion of infrared detection tasks.