Vehicle and Pedestrian Detection Method for Fusion Image Based on MDA-YOLOv10

To address the challenges of large-scale variations, difficulty in detecting small objects, and significant background interference in complex traffic scenarios, this study proposes a vehicle and pedestrian detection algorithm based on infrared and visible image fusion using MDA-YOLOv10. Firstly, the MPDA module was designed to capture multi-scale information through attention heads with different dilation rates, enhancing the model's ability to extract multi-scale features. Secondly, the DASI module integrates multi-level feature fusion and an adaptive selection mechanism, which selectively retains target feature information to improve small target detection performance. Finally, the ATFL loss function is introduced to optimize the original loss function, effectively filtering background noise and enabling the model to better focus on target features in complex backgrounds. Experimental results show that MDA-YOLOv10 achieves mAP@0.5 of 84.4% and 81.6% on the M3FD and MSRS datasets, respectively, representing improvements of 3.8% and 4.8% compared to the original YOLOv10 algorithm. This demonstrates that the proposed method delivers superior detection performance in complex traffic scenarios.