A Real-Time and Long-Term Face Tracking Method Using Convolutional Neural Network and Optical Flow in IoT-Based Multimedia Communication Systems

Table 3

Stability values with respect to Equation (3) for MDNet, MTCNN, and C-OF on aforementioned four videos with different conditions.

Method	Active camera
	Clip 1	Clip 2	Clip 3	Clip 4	Clip 5
MDNet	4.106844	4.387077	5.088164	5.704352	4.977256
MTCNN	6.679212	6.522255	6.166027	8.261213	9.037096
C-OF (ours)	2.936315	3.350025	3.372295	3.907079	3.864539
Method	Active human
	Clip 1	Clip 2	Clip 3	Clip 4	Clip 5
MDNet	3.045595	2.81129	3.305667	3.098207	3.322302
MTCNN	4.064902	3.55413	3.513855	4.050621	3.746015
C-OF (ours)	0.851283	0.993135	0.728653	1.208135	1.403328
Method	Static human
	Clip 1	Clip 2	Clip 3	Clip 4	Clip 5
MDNet	1.425355	1.47768	1.453236	1.518395	2.260055
MTCNN	1.994141	2.51177	2.318275	2.333235	2.237117
C-OF (ours)	0.310882	0.308686	0.295996	0.245097	0.321575
Method	Active illumination
	Clip 1	Clip 2	Clip 3	Clip 4	Clip 5
MDNet	1.961921	2.130076	2.3661	3.036336	2.243076
MTCNN	2.831262	2.758897	3.116406	4.073383	2.963807
C-OF (ours)	0.452622	0.579371	0.696979	1.006401	0.392867