Aiming at the phenomena of leakage, false detection and low detection accuracy in complex scenes such as dense objects, small objects in distant view, etc., which occur in the helmet detection algorithm of two-wheeled vehicles, an improved RT-DETR two-wheeled vehicle helmet detection algorithm is proposed on the basis of RT-DETR-r18. Firstly, a dual cross-stage multi-scale feature fusion module (DcspBlock) is designed, and a multi-core initialization module (PKIBlock) is integrated into the cross-stage module to enhance the ability of the backbone part to capture objects of different scales in the near and far scenes; secondly, a small object detection module Decoderhead-p2 is introduced into the Encoder part of RT-DETR to enhance the model"s accuracy in small object detection; finally, the original model"s GIOU is replaced by the improved loss function MPD_Focaler-IOU, and the adjustment parameters are set to reduce the impact of positive and negative sample imbalance on the model"s performance, and the minimum vertical distance is introduced to give a better performance in the fine localization of the bounding box. Experiments show that the improved RT-DETR model improves mAP50 and mAP50-95 by 3.6% and 3.7% on the TSHW dataset, respectively, and the amount of parameters is reduced by 17.6%, which effectively improves the performance of the two-wheeled vehicle helmet detection in complex scenes.