Table 2.

Counting performances of the selected models trained with Adam optimizer, 416 pixels of input image dimension and batches of 32 images.

	Video-1	Video-2	Video-3	Video-4	Video-5	Total
Ground truth	4	6	36	40	23	109
YOLOv8n
Output	4	4	36	22	23	89
True positives	3	4	32	19	20	78
False positives	1	0	4	3	3	11
False negatives	1	2	4	21	3	31
Correct count rate (%)	75.00	66.66	88.89	47.50	86.96	71.56
F-score	0.75	0.80	0.89	0.61	0.87	0.79
YOLOv8s
Output	3	4	36	39	24	106
True positives	3	4	33	31	22	93
False positives	0	0	3	8	2	13
False negatives	1	2	3	9	1	16
Correct count rate (%)	75.00	66.67	91.67	77.50	95.65	85.32
F-score	0.86	0.80	0.92	0.78	0.94	0.87
YOLOv8m
Output	3	4	36	33	22	98
True positives	3	4	32	27	21	87
False positives	0	0	4	6	1	11
False negatives	1	2	4	13	2	22
Correct count rate (%)	75.00	66.67	88.89	67.50	91.30	79.82
F-score	0.86	0.80	0.89	0.74	0.93	0.84
YOLOv7-Tiny
Output	4	5	41	40	24	114
True positives	3	4	34	30	20	91
False positives	1	1	7	10	4	23
False negatives	1	2	2	10	3	18
Correct count rate (%)	75.00	66.67	94.44	75.00	86.96	83.49
F-score	0.75	0.73	0.88	0.75	0.85	0.82