Table 5. Compare the detection performance, detection speed, number of parameters of YOLOv4 and the proposed model (YOLOv4-ours) on the VisDrone2019 dataset.

Model	Class	P/%	R/%	mAP/%	Parameter/million	t/s
YOLOv4	All	53.91	42.88	42.79	63.99	0.025
	Pedestrian	61.66	47.57	50.48
	People	48.67	31.04	30.76
	Bicycle	44.01	19.16	19.57
	Car	70.30	75.44	76.42
	Van	57.58	46.12	48.62
	Truck	58.49	52.28	52.13
	Tricycle	45.89	29.05	26.40
	Awning-tricycle	32.98	23.07	18.78
	Bus	64.45	61.24	61.72
	Motor	55.08	43.79	42.98
YOLOv4-ours	All	62.71	51.20	52.76	44.79	0.028
	Pedestrian	68.98	56.04	60.59
	People	58.45	36.89	39.14
	Bicycle	50.72	29.42	29.81
	Car	77.16	82.41	83.72
	Van	65.84	52.63	56.59
	Truck	68.11	59.57	62.95
	Tricycle	55.14	39.41	38.19
	Awning-tricycle	44.53	33.78	30.92
	Bus	73.73	70.23	72.56
	Motor	64.48	51.58	53.15