Real-Time Object Detection for the Running Train Based on the Improved YOLO V4 Neural Network

Table 3

Comparison of AP values of different network models for four types of objects.

Detection algorithm		Track (%)	Signal (%)	Tunnel (%)	Train (%)
Faster R–CNN		97.82	84.87	85.68	91.73
SSD		96.59	73.62	74.60	92.01
YOLO V3		97.32	86.37	87.41	92.05
YOLO V4		97.99	89.78	89.03	92.77
YOLO V4-tiny		97.58	86.34	87.29	90.16
MYOLO-lite series	MV1-YOLO-lite	99.99	92.92	97.28	92.29
	MV2-YOLO-lite	99.41	87.87	93.73	95.68
	MV3-YOLO-lite	99.81	91.59	96.26	95.30