Adaptive Feature Fusion Object Tracking Based on Circulant Structure with Kernel

Gaopeng Zhao; Yupeng Shen; Jianyu Wang

doi:10.3788/AOS201737.0815001

Journals >Acta Optica Sinica >Volume 37 >Issue 8 >Page 0815001 > Article

Acta Optica Sinica
Vol. 37, Issue 8, 0815001 (2017)

Adaptive Feature Fusion Object Tracking Based on Circulant Structure with Kernel

Gaopeng Zhao^*, Yupeng Shen, and Jianyu Wang

Author Affiliations

Department of Automation, Nanjing University of Science and Technology, Nanjing, Jiangsu 210094, China

show less

DOI: 10.3788/AOS201737.0815001 Cite this Article Set citation alerts

Gaopeng Zhao, Yupeng Shen, Jianyu Wang. Adaptive Feature Fusion Object Tracking Based on Circulant Structure with Kernel[J]. Acta Optica Sinica, 2017, 37(8): 0815001 Copy Citation Text

show less

Fig. 1. Tracking results of CSK algorithm. (a) Illumination variation; (b) occlusion; (c) scale variation

Download full size

Fig. 2. Tracking results of the proposed algorithm and CSK algorithm. (a) Fish; (b) car24; (c) suv

Download full size

Fig. 3. Center location error comparison of the proposed algorithm and CSK algorithm. (a) Fish; (b) car24; (c) suv

Download full size

Fig. 4. Tracking results of different tracking algorithms. (a) Bird2; (b) bolt2; (c) human8; (d) jogging2; (e) car1; (f) walking2

Download full size

Fig. 5. Distance precision of tracking algorithms. (a) Bird2; (b) bolt2; (c) human8; (d) jogging2; (e) car1; (f) walking2

Download full size

Fig. 6. Success rate of tracking algorithms. (a) Bird2; (b) bolt2; (c) human8; (d) jogging2; (e) car1; (f) walking2

Download full size

Sequence	Characteristic	Sequence	Characteristic
Car24	Illumination variation	Human8	Illumination variation, scale variation
Fish	Illumination variation	Jogging2	Occlusion
Suv	Occlusion	Car1	Illumination variation, scale variation
Bird2	Occlusion, deformation	Walking2	Scale variation, occlusion
Bolt2	Deformation, background clutters

Table 1. Characteristics of video sequences in the experiment

Sequence	Algorithm
Sequence	KCF	STRUCK	TLD	MIL	CT	DSST	Proposed algorithm
Bird2	0.474	0.545	$\begin{matrix} \bar{0.869} \end{matrix}$	0.646	0.101	0.474	0.979
Bolt2	0.017	0.109	0.013	1.000	0.638	0.020	$\begin{matrix} \bar{0.993} \end{matrix}$
Human8	1.000	$\begin{matrix} \bar{0.195} \end{matrix}$	0.187	0.156	0.078	1.000	1.000
Jogging2	0.162	0.254	$\begin{matrix} \bar{0.856} \end{matrix}$	0.182	0.166	0.185	1.000
Car1	$\begin{matrix} \bar{0.738} \end{matrix}$	0.330	0.591	0.239	0.141	1.000	1.000
Walking2	0.438	$\begin{matrix} \bar{0.966} \end{matrix}$	0.424	0.406	0.432	1.000	1.000

Table 2. Distance precision of different tracking algorithms

Sequence	Algorithm
Sequence	KCF	STRUCK	TLD	MIL	CT	DSST	Proposed algorithm
Bird2	0.474	0.525	$\begin{matrix} \bar{0.707} \end{matrix}$	0.595	0.101	0.474	0.979
Bolt2	0.006	0.006	0.006	0.965	0.419	0.010	$\begin{matrix} \bar{0.686} \end{matrix}$
Human8	$\begin{matrix} \bar{0.304} \end{matrix}$	0.132	0.132	0.156	0.007	1.000	1.000
Jogging2	0.159	0.162	0.156	0.162	0.140	$\begin{matrix} \bar{0.182} \end{matrix}$	0.996
Car1	0.053	0.053	0.293	0.053	0.007	$\begin{matrix} \bar{0.604} \end{matrix}$	1.000
Walking2	0.378	$\begin{matrix} \bar{0.434} \end{matrix}$	0.338	0.380	0.384	1.000	1.000

Table 3. Success rate of different tracking algorithms

Algorithm	Sequence
Algorithm	Bird2	Bolt2	Human8	Jogging	Car1	Walking2
CSK	51.4	130.5	139.1	81.1	175.1	80.7
DSST	7.2	21.6	17.5	10.1	18.0	12.5
Proposed algorithm	20.0	39.0	34.0	28.7	49.6	30.9

Table 4. Complexity analysisframe·s-1

Gaopeng Zhao, Yupeng Shen, Jianyu Wang. Adaptive Feature Fusion Object Tracking Based on Circulant Structure with Kernel[J]. Acta Optica Sinica, 2017, 37(8): 0815001

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information