Multiscale Traffic Sign Detection Method in Complex Environment Based on YOLOv4

Table 1

Comparison of proposed and other methods.

Model	mAPNoChallenge	mAPRain	mAPSnow	mAPFog	mAPLensBlur	mAP	FPS
Faster-RCNN	88.86	70.51	71.07	69.28	73.05	82.59	24
SSD	78.71	61.54	63.56	60.97	64.58	72.12	51
RetinaNet	78.07	59.55	60.91	63.89	64.76	71.84	53
YOLOv3	80.18	63.94	63.83	62.81	66.25	74.32	58
YOLOv4-tiny	79.11	61.57	61.45	60.32	63.59	72.03	142
YOLOv4	83.92	66.62	67.14	65.59	69.26	77.25	86
Ours	87.19	73.22	74.02	72.50	76.41	81.78	74