Object Tracking Algorithm Based on Correlation Filtering and Convolution Residuals Learning

Yaguang Yang; Zhenhong Shang

doi:10.3788/LOP57.121012

Journals >Laser & Optoelectronics Progress >Volume 57 >Issue 12 >Page 121012 > Article

Laser & Optoelectronics Progress
Vol. 57, Issue 12, 121012 (2020)

Object Tracking Algorithm Based on Correlation Filtering and Convolution Residuals Learning

Yaguang Yang^** and Zhenhong Shang^*

Author Affiliations

Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming, Yunnan 650500, China

show less

DOI: 10.3788/LOP57.121012 Cite this Article Set citation alerts

Yaguang Yang, Zhenhong Shang. Object Tracking Algorithm Based on Correlation Filtering and Convolution Residuals Learning[J]. Laser & Optoelectronics Progress, 2020, 57(12): 121012 Copy Citation Text

show less

Fig. 1. Network frame diagram of proposed algorithm

Download full size

Fig. 2. Basic convolution layer and temporal-spatial residual layer

Download full size

Fig. 3. Comparison experiment on OTB2015 datasets with baseline. (a) Precision; (b) success rate

Download full size

Fig. 4. Distance precision plots and overlap success plots of ten algorithms in OTB-2013. (a) Precision; (b) success rate

Download full size

Fig. 5. Distance precision plots and overlap success plots of ten algorithms in OTB-2015. (a) Precision; (b) success rate

Download full size

Fig. 6. A visualization of the tracking results of seven algorithms on eight sequences

Download full size

Item	Proposed	CF2	Scale_DLSSVM	DeepSRDCF	SiamFC	Staple	SAMF	KCF	DSST	Struck
IV	0.862	$\begin{matrix} \bar{0.816} \end{matrix}$	0.790	0.786	0.741	0.787	0.708	0.724	0.715	0.558
SV	0.872	0.798	0.758	$\begin{matrix} \bar{0.817} \end{matrix}$	0.738	0.723	0.701	0.635	0.633	0.595
OCC	0.893	0.765	0.789	$\begin{matrix} \bar{0.822} \end{matrix}$	0.726	0.721	0.722	0.632	0.590	0.528
DEF	0.849	$\begin{matrix} \bar{0.790} \end{matrix}$	0.748	0.779	0.693	0.743	0.680	0.619	0.533	0.527
MB	0.883	0.804	0.740	$\begin{matrix} \bar{0.823} \end{matrix}$	0.705	0.707	0.655	0.601	0.567	0.580
FM	0.871	$\begin{matrix} \bar{0.815} \end{matrix}$	0.727	0.814	0.743	0.697	0.654	0.621	0.552	0.606
IPR	0.860	$\begin{matrix} \bar{0.854} \end{matrix}$	0.820	0.818	0.742	0.770	0.721	0.701	0.691	0.629
OPR	0.886	$\begin{matrix} \bar{0.807} \end{matrix}$	0.802	0.835	0.756	0.738	0.739	0.677	0.644	0.587
OV	0.864	0.677	0.704	$\begin{matrix} \bar{0.781} \end{matrix}$	0.669	0.661	0.628	0.501	0.481	0.472
BC	0.866	$\begin{matrix} \bar{0.843} \end{matrix}$	0.793	0.841	0.690	0.766	0.689	0.713	0.704	0.552
LR	0.872	0.831	0.791	0.708	$\begin{matrix} \bar{0.847} \end{matrix}$	0.631	0.685	0.560	0.567	0.671
Note: In this table, the number marked with black is the first, and the number underlined is the second.

Table 1. DP values of proposed method and compared methods under different scene attributes

Item	Proposed	CF2	Scale_DLSSVM	DeepSRDCF	SiamFC	Staple	SAMF	KCF	DSST	Struck
IV	0.677	0.541	0.572	$\begin{matrix} \bar{0.624} \end{matrix}$	0.574	0.596	0.530	0.482	0.556	0.428
SV	0.652	0.485	0.500	$\begin{matrix} \bar{0.607} \end{matrix}$	0.556	0.522	0.492	0.395	0.466	0.403
OCC	0.672	0.526	0.551	$\begin{matrix} \bar{0.603} \end{matrix}$	0.547	0.545	0.538	0.445	0.449	0.393
DEF	0.615	0.530	0.522	$\begin{matrix} \bar{0.567} \end{matrix}$	0.510	0.552	0.505	0.438	0.415	0.387
MB	0.701	0.585	0.591	$\begin{matrix} \bar{0.642} \end{matrix}$	0.550	0.546	0.525	0.459	0.469	0.459
FM	0.673	0.570	0.550	$\begin{matrix} \bar{0.628} \end{matrix}$	0.568	0.537	0.507	0.459	0.447	0.462
IPR	0.622	0.559	0.564	$\begin{matrix} \bar{0.589} \end{matrix}$	0.557	0.552	0.519	0.469	0.502	0.448
OPR	0.648	0.534	0.547	$\begin{matrix} \bar{0.607} \end{matrix}$	0.558	0.534	0.536	0.453	0.470	0.423
OV	0.639	0.474	0.498	$\begin{matrix} \bar{0.553} \end{matrix}$	0.506	0.481	0.480	0.393	0.386	0.365
BC	0.645	0.585	0.560	$\begin{matrix} \bar{0.627} \end{matrix}$	0.523	0.574	0.525	0.498	0.523	0.429
LR	0.597	0.439	0.433	0.475	$\begin{matrix} \bar{0.592} \end{matrix}$	0.418	0.430	0.307	0.383	0.363
Note: In this table, the number marked with black is the first, and the number underlined is the second.

Table 2. OP values of proposed method and compared methods under different scene attributes

Yaguang Yang, Zhenhong Shang. Object Tracking Algorithm Based on Correlation Filtering and Convolution Residuals Learning[J]. Laser & Optoelectronics Progress, 2020, 57(12): 121012

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information