Aiming at the problem that small targets are prone to misdetection and missing detection in vehicle pedestrian detection, a vehicle pedestrian detection algorithm based on improved YOLOv5s is proposed. First, a fourth detection head is extracted from the head network without changing the original PANet to improve the detection ability of small targets. Secondly, CF module is designed in the backbone network to replace the original subsampling module and enhance the feature extraction ability. Then C3 module, which has half the number of channels in the Neck network, is redesigned into S-C3 to reduce the information loss problem in the Neck part. Finally, SPPF module is rebuilt into D-SPP module to retain more detailed features for the Neck part. Improve the detection accuracy of the target. KITTI data set was used in the experiment to merge and delete the categories of the data set. The experimental results show that the improved algorithm is better than the original algorithm on KITTI for the problem of misdetection and missing detection of small targets. And still meet the requirements of real-time.