Object detection is a crucial component in achieving autonomous driving technology. How to detect the object quickly and accurately in the complex scene is challenging for autonomous driving system. To solve the above problem, an improved You Only Look Once Version 5s algorithm is proposed. This paper focuses on enhancing the detection capabilities for small objects by introducing an improved feature layer structure. By adding an additional small object detection layer and designing feature fusion modules, the algorithm effectively aggregates multi-layer features, thereby improving the detection accuracy of small objects. Then, a comparative experiment is conducted on the KITTI dataset. The results demonstrate that the proposed algorithm achieves better detection results and achieves a balance between detection accuracy and speed.