An enhanced YOLO detection model is proposed for the small objects in traffic scenes with the characteristics of small detection object scale and the problems of misdetection and missed detection that are easy to occur in the mainstream use methods. Within the backbone network, we have incorporated a specialized convolutional block adept at detecting small objects, replacing the standard convolutional module from the initial network, thereby enhancing the detection performance for small scale objects; the channel attention mechanism is incorporated in the network’s neck section to improve feature extraction efficiency without introducing extra computational burden; a novel detection head designed for small objects has been incorporated to accommodate targets of varying scales and to boost the accuracy of detecting small scale objects; finally, the inclusion of a loss function based on auxiliary edges is incorporated to expedite model convergence and increase generalization capacity. The experimental findings from both our exclusive datasets and the Visdrone2019 dataset highlight the algorithm’s dominance compared to its counterparts, thereby robustly confirming the model’s effectiveness and its ability to generalize.