Adaptive Correlation Filtering Tracking Algorithm for Complex Scenes

Mingrui Lu; Chao Han; Fan Lu; Baorui Miao; Jikun Yang; Junjun Zha; Wenhan Sha

doi:10.3788/LOP202259.2415004

Journals >Laser & Optoelectronics Progress >Volume 59 >Issue 24 >Page 2415004 > Article

Laser & Optoelectronics Progress
Vol. 59, Issue 24, 2415004 (2022)

Adaptive Correlation Filtering Tracking Algorithm for Complex Scenes

Mingrui Lu^1、2, Chao Han^1、2、*, Fan Lu^1、2, Baorui Miao^1、2, Jikun Yang^1、2, Junjun Zha^1、2, and Wenhan Sha³

Author Affiliations

¹School of Electrical Engineering, Anhui Polytechnic University, Wuhu 241000, Anhui , China

²Key Laboratory of Advanced Perception and Intelligent Control of High-End Equipment, Ministry of Education, Anhui Polytechnic University, Wuhu 241000, Anhui , China

³Chery New Energy Automobile Co., Ltd., Wuhu 241000, Anhui , China

show less

DOI: 10.3788/LOP202259.2415004 Cite this Article Set citation alerts

Mingrui Lu, Chao Han, Fan Lu, Baorui Miao, Jikun Yang, Junjun Zha, Wenhan Sha. Adaptive Correlation Filtering Tracking Algorithm for Complex Scenes[J]. Laser & Optoelectronics Progress, 2022, 59(24): 2415004 Copy Citation Text

show less

Response diagrams corresponding to Joggle-1 sequence. (a) No occlusion image in frame 4; (b) response diagram of frame 4;(c) occlusion image in frame 49; (d) response diagram of frame 49

Fig. 1. Response diagrams corresponding to Joggle-1 sequence. (a) No occlusion image in frame 4; (b) response diagram of frame 4;(c) occlusion image in frame 49; (d) response diagram of frame 49

Download full size

Fig. 2. Flowchart of the proposed algorithm

Download full size

Fig. 3. Influence of different parameters on tracker performance. (a) Parameter ζ; (b) parameter τ; (c) parameters η1 and η2

Download full size

Fig. 4. Precision and success rate of different feature weighted methods. (a) Precision; (b) success rate

Download full size

Fig. 5. Scale changes of different algorithms under three groups of different video sequences. (a) Blurcar2; (b) Doll; (c) Carscale

Download full size

Fig. 6. ΔEAPC value and changes of each frame under the Joggle sequences. (a) Joggle-1; (b) Joggle-2

Download full size

Fig. 7. CLE changing between two sets of videos. (a) Basketball; (b) Faceocc1

Download full size

Fig. 8. Precision and success rate of seven algorithms on OTB50 dataset. (a) Precision; (b) success rate

Download full size

Fig. 9. Precision and success rate of seven algorithms on OTB2015 dataset. (a) Precision; (b) success rate

Download full size

Fig. 10. Comparison of seven algorithms on different video sequences. (a) Box; (b) Dragonbaby; (c) Bird2; (d) Panda; (e) Carscale; (f) Soccer; (g) Tiger2

Download full size

Parameter	$α$	$β$	$γ$
$λ$ ₁	0.5	0.4	0.1
$λ$ ₂	0.3	0.6	0.1
$λ$ ₃	1/3	1/3	1/3

Table 1. Three weighted ratios of multi-feature fusion

Sequence	Precision	Success rate	Speed /（frame·s^-1）
Blurcar2	0.999	1.000	44.06
Carscale	0.897	0.980	60.42
Doll	0.961	0.966	57.55

Table 2. Index of the proposed algorithm under three groups of scale sequences

Parameter	OURS	OURS4	OURS5	OURS6	SRDCF	SRDCFdecon	SAMF
Success rate	0.674	0.564	0.626	0.656	0.634	0.677	0.576
Speed /（frame·s^-1）	39.07	52.10	44.24	35.52	2.05	5.18	14.54

Table 3. Result comparison of search area algorithms

Algorithm	IV	DEF	SV	OCC	MB	FM	IPR	OPR	OV	BC	LR
OURS	0.760	0.741	0.755	0.764	0.787	0.732	0.805	0.790	0.630	0.808	0.707
SRDCFdecon	0.803	0.730	0.771	0.734	0.808	0.763	0.715	0.759	0.561	0.814	0.600
SRDCF	0.715	0.693	0.689	0.673	0.734	0.744	0.631	0.675	0.520	0.676	0.594
SAMF	0.648	0.650	0.663	0.706	0.674	0.657	0.665	0.701	0.605	0.630	0.645
fDSST	0.715	0.589	0.628	0.612	0.658	0.660	0.680	0.617	0.476	0.720	0.609
DSST	0.677	0.509	0.578	0.552	0.585	0.548	0.644	0.598	0.357	0.648	0.510
KCF	0.676	0.573	0.591	0.603	0.622	0.614	0.651	0.624	0.428	0.677	0.538

Table 4. Precision of seven target tracking algorithms at 11 different challenge attributes

Algorithm	IV	DEF	SV	OCC	MB	FM	IPR	OPR	OV	BC	LR
OURS	0.692	0.667	0.633	0.687	0.742	0.683	0.688	0.693	0.531	0.720	0.501
SRDCFdecon	0.748	0.641	0.712	0.694	0.795	0.729	0.649	0.686	0.561	0.740	0.571
SRDCF	0.675	0.626	0.631	0.640	0.719	0.712	0.574	0.607	0.492	0.634	0.581
SAMF	0.584	0.555	0.562	0.641	0.660	0.598	0.602	0.622	0.490	0.596	0.539
fDSST	0.645	0.525	0.547	0.553	0.633	0.641	0.610	0.545	0.438	0.645	0.547
DSST	0.601	0.437	0.489	0.491	0.568	0.515	0.552	0.506	0.305	0.535	0.423
KCF	0.503	0.434	0.396	0.475	0.565	0.534	0.515	0.477	0.376	0.576	0.325

Table 5. Success rate of seven target tracking algorithms at 11 different challenge attributes

Algorithm	Speed （frame·s^-1）
Algorithm	OTB50	OTB2015
OURS	39.0778	39.2448
SRDCFdecon	2.0497	1.9694
SRDCF	5.1833	4.871
SAMF	14.5473	13.933
fDSST	37.4683	31.6959
DSST	8.4053	7.0361
KCF	198.4457	173.5403

Table 6. Running speed of seven target tracking algorithms

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information