<p>Object detection is a critical component of environmental perception in advanced driver assistance systems (ADAS). Despite the rapid development of object detection algorithms, significant challenges remain in practical applications for small objects (such as people, bicycles, and traffic signs). These challenges include occlusion, low resolution, multi-scale targets, and complex dynamic scenes. To address these issues, this paper proposes a novel small object detection algorithm called SBEW-YOLOv8. Firstly, the backbone network incorporates Space-to-Depth Convolution (SPDConv), which is suitable for low-resolution images and small object detection. This technique preserves discriminative feature information and enhances the perception of small objects. Secondly, an efficient multi-scale feature fusion structure, small object bidirectional feature pyramid network (SO-BiFPN), is designed in the neck network to enhance the information exchange between feature layers of different scales. Additionally, efficient multi-scale attention (EMA) is employed to highlight target features and suppress background interference. Finally, the wise intersection over union (WIoU) loss function with adaptive weight adjustment is used to improve the robustness of detection and the regression performance of bounding boxes. This algorithm was tested in various autonomous driving scenarios, including traffic light intersections, rural open roads, curved intersections, and low-light intersections. On the publicly available SODA-D dataset, the average precision for small object detection reached 34.8%, an improvement of 4.6% compared to the baseline model YOLOv8. The results demonstrate that SBEW-YOLOv8 significantly enhances the model’s ability to detect small objects.</p>