An Improvement of the Camshift Human Tracking Algorithm Based on Deep Learning and the Kalman Filter

Table 1

The results of error calculation of the parameters of state vectors.

Condition	Case	Frame
Good lighting room	Multiple people using the same shirt	2	2	−3	−3	−6	−1	−2
		15	1	0	3	1	0	−3
		30	0	1	1	−2	−1	−1
		35	−27	−18	−16	−59	−2	2
		45	−16	6	5	−5	1	0
	Multiple people using the different shirt	2	0	5	3	1	0	−5
		15	−1	2	0	−2	−1	3
		25	14	−11	7	13	4	0
		30	23	2	13	8	−3	0
		35	2	0	1	−3	0	−1
Poor lighting room	Multiple people using the same shirt	2	−1	−3	−1	−5	1	−1
		30	−2	−3	0	−2	0	1
		120	−1	−4	1	−6	−1	0
		130	5	−6	5	−8	1	−2
		160	60	4	−2	−48	0	1
	Multiple people using the different shirt	2	0	1	−1	−11	0	0
		20	3	1	0	−8	−1	−2
		25	34	−3	−2	−16	0	3
		30	1	2	−3	−10	−3	1
		35	0	3	−1	−9	0	1