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

Table 4

Smoothness values with respect to Equation (5) for MDNet, MTCNN, and C-OF on four videos with different conditions.

Method	Active camera
	Clip 1	Clip 2	Clip 3	Clip 4	Clip 5
MDNet	0.152312	0.146476	0.303847	0.270567	0.088137
MTCNN	1.071629	0.683341	0.650429	0.581538	0.656829
C-OF (ours)	0.005137	0.011361	0.010636	0.005119	0.009432
Method	Active human
	Clip 1	Clip 2	Clip 3	Clip 4	Clip 5
MDNet	0.678967	0.491966	0.586164	0.699405	0.402297
MTCNN	1.145627	0.847167	0.93854	0.977966	0.971969
C-OF (ours)	0.001103	0.003444	0.001304	0.004872	0.021336
Method	Static human
	Clip 1	Clip 2	Clip 3	Clip 4	Clip 5
MDNet	0.393428	0.43518	0.444789	0.521464	0.612867
MTCNN	0.5484	0.673217	0.729767	0.625556	0.566896
C-OF (ours)	0.004908	0.000072	0.0	0.000008	0.0
Method	Active illumination
	Clip 1	Clip 2	Clip 3	Clip 4	Clip 5
MDNet	0.417814	0.3991	0.564509	0.531881	0.48117
MTCNN	1.020253	0.616373	0.805697	1.110985	1.045736
C-OF (ours)	0.010921	0.000995	0.0	0.007117	0.000901