Scale Adaptive Kernel Correlation Tracking Method with High Confidence

Fujin Li; Huihui Liu; Hongge Ren; Tao Shi

doi:10.3788/LOP202158.0815004

Journals >Laser & Optoelectronics Progress >Volume 58 >Issue 8 >Page 0815004 > Article

Laser & Optoelectronics Progress
Vol. 58, Issue 8, 0815004 (2021)

Scale Adaptive Kernel Correlation Tracking Method with High Confidence

Fujin Li, Huihui Liu^*, Hongge Ren, and Tao Shi

Author Affiliations

College of Electrical Engineering, North China University of Science and Technology, Tangshan, Hebei 063210, China

show less

DOI: 10.3788/LOP202158.0815004 Cite this Article Set citation alerts

Fujin Li, Huihui Liu, Hongge Ren, Tao Shi. Scale Adaptive Kernel Correlation Tracking Method with High Confidence[J]. Laser & Optoelectronics Progress, 2021, 58(8): 0815004 Copy Citation Text

show less

Fig. 1. Flowchart of tracing method

Download full size

Fig. 2. Precision curves of different methods in 100 sequences

Download full size

Fig. 3. Success rate curves of different methods in 100 sequences

Download full size

Fig. 4. Qualitative comparison results of six trackers in different sequences. (a) shaking sequence; (b) freeman1 sequence; (c) deer sequence

Download full size

Name	Translation confidence criterion	Scale confidence criterion	Value
Ours 1	$\{\begin{array}{l} A_{APCE, p, t} > θ \\ \max [S_{p, t}] > θ \end{array}$	$\{\begin{array}{l} A_{APCE, scale, t} > θ \\ \max [S_{scale, t}] > θ \end{array}$	θ=0.5
Ours 2	$\{\begin{array}{l} A_{APCE, p, t} > α_{1} \frac{1}{t - 1} \overset{t - 1}{\sum_{i = 1}} A_{APCE, p, t}^{(i)} \\ \max [S_{p, t}] > α_{2} \frac{1}{t - 1} \overset{t - 1}{\sum_{i = 1}} \max [S_{p, t}^{(i)}] \end{array}$	$\{\begin{array}{l} A_{APCE, scale, t} > α_{1} \frac{1}{t - 1} \overset{t - 1}{\sum_{i = 1}} A_{APCE, scale, t}^{(i)} \\ \max [S_{scale, t}] > α_{2} \frac{1}{t - 1} \overset{t - 1}{\sum_{i = 1}} \max [S_{scale, t}^{(i)}] \end{array}$	$\begin{array}{l} α_{1} = 0.45 \\ α_{2} = 0.70 \end{array}$

Table 1. Threshold setting for tracking confidence

Paramater	Value
CF (HOG)model learning rate	0.01
Background and foreground colour models learning rate	0.04
Scale learning rate	0.008
Interpolation factor	0.3

Table 2. Setting of experimental parameters

Parameter	Ours 1	Ours 2	SAMF	KCC	KCF	DCF
Average precision	0.794	0.761	0.751	0.755	0.696	0.690
Average success rate	0.694	0.673	0.674	0.565	0.551	0.548
Average frame rate/(frame×s^-1)	15.340	16.350	5.476	16.680	72.382	105.650

Table 3. Performance comparison of different trackers

Attribute	Ours 1	Ours 2	SAMF	KCC	KCF	DCF
SV(63)	0.745	0.715	0.701	0.701	0.635	0.628
FM(39)	0.714	0.680	0.654	0.711	0.621	0.603
BC(31)	0.832	0.814	0.689	0.808	0.713	0.686
MB(29)	0.700	0.691	0.655	0.713	0.601	0.576
IV(37)	0.809	0.790	0.708	0.808	0.724	0.698
OV(14)	0.682	0.610	0.628	0.652	0.501	0.487
IPR(51)	0.802	0.782	0.721	0.750	0.701	0.686
DEF(43)	0.756	0.693	0.680	0.751	0.619	0.623
LR(9)	0.606	0.518	0.685	0.494	0.560	0.564
OCC (48)	0.724	0.685	0.722	0.679	0.632	0.611
OPR (63)	0.754	0.715	0.739	0.716	0.677	0.665

Table 4. Average precision facing different tracking challenges

Attribute	Ours 1	Ours 2	SAMF	KCC	KCF	DCF
SV(63)	0.606	0.589	0.584	0.495	0.415	0.416
FM(39)	0.640	0.630	0.595	0.599	0.526	0.521
BC(31)	0.783	0.769	0.639	0.625	0.609	0.597
MB(29)	0.640	0.653	0.641	0.651	0.550	0.542
IV(37)	0.707	0.707	0.640	0.599	0.550	0.534
OV(14)	0.603	0.566	0.551	0.516	0.457	0.450
IPR(51)	0.696	0.680	0.641	0.548	0.553	0.544
DEF(43)	0.651	0.608	0.606	0.606	0.503	0.505
LR(9)	0.418	0.382	0.515	0.244	0.295	0.301
OCC (48)	0.653	0.630	0.664	0.530	0.512	0.500
OPR (63)	0.654	0.625	0.660	0.530	0.527	0.519

Table 5. Average success rate facing different tracking challenges

Fujin Li, Huihui Liu, Hongge Ren, Tao Shi. Scale Adaptive Kernel Correlation Tracking Method with High Confidence[J]. Laser & Optoelectronics Progress, 2021, 58(8): 0815004

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information