Traffic safety is key to smart cities, but frequent accidents persist as a challenge. Surveillance systems generate massive real-time video data, yet edge devices’ computational limits hinder real-time processing, especially complex, computationally costly models reduce detection efficiency. To address these limitations, this study proposes an optimized lightweight model MX-YOLOv10, which integrates two key innovations based on DM-YOLOv10: using C3Ghost modules to replace the C2f modules in the neck network; employing Reparameterized Convolution modules to replace the standard Conv-BN-SiLU blocks in the backbone. Experimental results demonstrate that compared to DM-YOLOv10, the proposed algorithm achieves 104.2 FPS, a significant increase from the baseline of 70.4 FPS, while maintaining a mAP@50 of 88.52%. Furthermore, the model’s parameters decrease from 2.9 million to 2.5 million, and computational costs reduce from 8.7 GFLOPs to 6.8 GFLOPs. These advancements position MX-YOLOv10 as an efficient solution for real-time traffic monitoring on resource-constrained edge devices.