Efficient Eye-Blinking Detection on Smartphones: A Hybrid Approach Based on Deep Learning

Table 2

Detailed performance comparison of two ROI adjustment methods.

Experiment ID	Number of frames in tracking phase	Number of frames with eyes undetected (in current ROI)	Number of successes (i.e., eye is found after ROI adjustment)	Prediction success rate (%)
(a) Regression-based prediction
1	1320	368	88	24
2	1884	452	128	44
3	1384	264	108	41
4	3021	130	56	43
5	3702	342	101	30
(b) Naïve version of [7]
1	972	226	42	9
2	404	276	60	22
3	1452	204	42	21
4	3043	153	31	20
5	3421	311	53	17