Table 2.

Multiple class identification accuracy and agreement results for all videos between human annotators and the computer vision algorithm using the Re-ID technique

Video	Duration of identity - manual (s)		Duration of identity – algorithm (s)		Δ (s)	Normalized Δ (%)	Cohen’s kappa	Re-ID accuracy (%)
	Patient Caregiver	Researchers	Patient Caregiver	Researchers
#1	30.16	0.28	30.44	0.16	– 0.16	0.27	0.93	96.26
#2	26.90	0.10	26.5	0.10	0.40	0.66	0.88	94.18
#3	28.76	1.32	30.16	1.24	– 1.32	2.20	0.92	95.94
#4	34.20	1.16	35.72	0.84	– 1.20	1.98	0.93	96.50
#5	24.80	0.00	26.00	0.00	– 1.20	1.99	0.96	98.01
#6	24.56	0.08	26.56	0.12	– 2.04	3.40	0.91	95.80
#7	24.00 10.36	0.04	23.68 10.80	0.08	– 0.16	0.27	0.91	94.53

Note: Duration of identity: the time (in s) a specific individual is identified. The identities are patients and researchers, in video #7 additionally a caregiver is identified, the time (in s) the caregiver is identified is shown in italics