Table 1.

Comparative experimental results.

Methods	P	R	F1	AP	FPS
Faster-R-CNN	70.13%	63.20%	0.6648	68.75%	27.26
Cascade R-CNN	78.32%	68.08%	0.7284	74.68%	8.63
YOLOv3	78.51%	67.98%	0.7286	73.85%	17.21
YOLOv3+ multiple-frame input	87.59%	75.72%	0.8122	79.43%	6.30
YOLOv3+ multiple-frame input+ optical flow	89.21%	78.36%	0.8343	85.18%	0.87
YOLOv4	83.47%	79.15%	0.8125	83.25%	20.05
YOLOv4+ multiple-frame input	89.28%	84.88%	0.8702	86.71%	10.17
YOLOv4+ multiple-frame input+ optical flow	89.68%	86.24%	0.8792	91.75%	1.48
YOLOv5x	85.92%	80.18%	0.8294	83.26%	20.21
YOLOv5l	83.41%	78.39%	0.8082	81.94%	31.14
YOLOv5m	81.56%	73.47%	0.7730	79.17%	52.63
YOLOv5s	78.77%	66.71%	0.7224	75.38%	66.45
YOLOv5s+ multiple-frame input	86.18%	76.43%	0.8101	82.27%	54.37
YOLOv5s+ multiple-frame input+ optical flow	86.96%	80.67%	0.8369	86.87%	31.84
YOLOv5n	76.38%	63.11%	0.6911	71.51%	75.34