改进的有雾图像中被遮挡车辆及行人识别算法

于天河,王文龙,刘镛,杨壮壮,侯善冲

Improved algorithm for identifying occluded vehicles and pedestrians in foggy images

Tianhe YU,Wenlong WANG,Yong LIU,Zhuangzhuang YANG,Shanchong HOU

表 5 经ABMD-Net去雾处理后的各模型检测性能量化对比表

Tab.5 Quantitative comparison of detection performance of each model after ABMD-Net defogging treatment

模型 mAP FPS/(帧·s−1) 薄雾 中雾 重雾 So Po Se ABMD-Net+SSD 0.7527 0.6956 0.5967 0.7645 0.5284 0.3763 9.74 ABMD-Net+Faster R-CNN 0.8075 0.7390 0.6258 0.7984 0.6842 0.5166 5.38 ABMD-Net+YOLOv10 0.8234 0.7812 0.6734 0.8026 0.7144 0.5428 32.79 AOD-Net+MsF-SSD-Net 0.8837 0.8365 0.7149 0.8332 0.7483 0.5556 25.03 ABMD-Net+MsF-SSD-Net 0.9192 0.8534 0.7378 0.8778 0.7634 0.5633 45.21 Defog YOLO[29] 0.8670 — — — — — — YOLOv5-Transformer[30] 0.8280 — — — — — — AO YOLO[31] 0.8910 — — — — — — ABMD-Net-MsF-SSD-Net 0.9385 0.8683 0.7824 0.8952 0.7893 0.5748 47.61