Multi-Filter Collaborative Tracking Algorithm Based on High-Confidence Updating Strategy

Chaoyi Zhang; Li Peng; Tianhao Jia; Jiwei Wen

doi:10.3788/LOP56.191502

Journals >Laser & Optoelectronics Progress >Volume 56 >Issue 19 >Page 191502 > Article

Laser & Optoelectronics Progress
Vol. 56, Issue 19, 191502 (2019)

Multi-Filter Collaborative Tracking Algorithm Based on High-Confidence Updating Strategy

Chaoyi Zhang, Li Peng, Tianhao Jia, and Jiwei Wen^*

Author Affiliations

Key Laboratory of Advanced Process Control for Light Industry, Ministry of Education, School of Internet of Things Engineering, Jiangnan University, Wuxi, Jiangsu 214122, China

show less

DOI: 10.3788/LOP56.191502 Cite this Article Set citation alerts

Chaoyi Zhang, Li Peng, Tianhao Jia, Jiwei Wen. Multi-Filter Collaborative Tracking Algorithm Based on High-Confidence Updating Strategy[J]. Laser & Optoelectronics Progress, 2019, 56(19): 191502 Copy Citation Text

show less

Fig. 1. Tracking results and response maps in Box sequence. (a)-(c) 199th, 462nd, and 505th frame images; (d)-(f) 199th, 462nd, and 505th frame response maps

Download full size

Fig. 2. Center location errors of two updating strategies in Box sequence

Download full size

Fig. 3. Specific steps of the proposed algorithm

Download full size

Fig. 4. Overall framework of the proposed algorithm

Download full size

Fig. 5. Results of nine tracking algorithms for nine sequences. (a) Lemming; (b) Human3; (c) Girl2; (d) DragonBaby; (e) MotorRolling; (f) Human7; (g) Skiing; (h) Shaking; (i) Tiger2

Download full size

Fig. 6. Overall performance of 9 tracking algorithms on OTB-100. (a) Distance precision curve; (b) success rate curve

Download full size

Fig. 7. Overall analysis of algorithm structure on OTB-100. (a) Distance precision curve; (b) success rate curve

Download full size

Fig. 8. Overall performance of our algorithm under different updating parameters on OTB-100. (a) Distance precision curve; (b) success rate curve

Download full size

Fig. 9. Overall performance of 9 tracking algorithms on TC-128. (a) Distance precision curve; (b) success rate curve

Download full size

Sequence	Attributes	Frame number
DragonBaby	SV, OCC, MB,FM, IPR, OPR, OV	113
Human3	SV, OCC, DEF, OPR, BC	1698
MotorRolling	IV, SV, MB, FM,IPR, BC, LR	164
Skiing	IV, SV, DEF, IPR, OPR	81
Shaking	IV,SV, IPR, OPR, BC	365
Tiger2	IV, OCC, DEF, MB,FM, IPR, OPR, OV	365
Girl2	SV, OCC, DEF, MB, OPR	1500
Human7	IV, SV, OCC, DEF, MB, FM	250
Lemming	IV, SV, OCC, FM, OPR, OV	1336

Table 1. Attributes and frame numbers of 9 video sequences

Algorithm	FM	BC	MB	DEF	IV	IPR	LR	OCC	OPR	OV	SV
Ours	0.822	0.882	0.833	0.821	0.866	0.855	0.870	0.815	0.850	0.751	0.830
HDT	0.817	0.844	0.789	0.817	0.820	0.844	0.887	0.774	0.805	0.663	0.808
HCF	0.815	0.843	0.804	0.791	0.831	0.864	0.847	0.778	0.816	0.677	0.799
CNN-SVM	0.747	0.776	0.751	0.793	0.795	0.813	0.925	0.730	0.798	0.650	0.787
SRDCF	0.769	0.775	0.767	0.734	0.792	0.745	0.765	0.735	0.742	0.597	0.745
Staple	0.710	0.749	0.699	0.751	0.782	0.768	0.695	0.728	0.738	0.668	0.727
LCT2	0.681	0.734	0.669	0.689	0.746	0.781	0.699	0.682	0.746	0.592	0.681
KCF	0.621	0.713	0.601	0.617	0.719	0.701	0.671	0.630	0.677	0.501	0.633
DSST	0.571	0.704	0.568	0.550	0.722	0.698	0.684	0.603	0.655	0.478	0.649

Table 2. Distance precision scores of nine algorithms in terms of 11 attributes

Algorithm	FM	BC	MB	DEF	IV	IPR	LR	OCC	OPR	OV	SV
Ours	0.622	0.637	0.648	0.583	0.638	0.597	0.489	0.605	0.601	0.572	0.583
HDT	0.568	0.578	0.574	0.543	0.535	0.555	0.401	0.528	0.533	0.472	0.486
HCF	0.570	0.585	0.585	0.530	0.549	0.566	0.388	0.532	0.540	0.474	0.485
CNN-SVM	0.546	0.548	0.578	0.547	0.537	0.548	0.379	0.515	0.548	0.488	0.490
SRDCF	0.597	0.583	0.594	0.544	0.613	0.544	0.514	0.559	0.550	0.460	0.561
Staple	0.541	0.561	0.540	0.550	0.593	0.548	0.399	0.542	0.533	0.476	0.520
LCT2	0.507	0.528	0.516	0.482	0.517	0.529	0.330	0.477	0.505	0.446	0.430
KCF	0.459	0.498	0.459	0.436	0.479	0.469	0.290	0.443	0.453	0.394	0.394
DSST	0.458	0.523	0.469	0.427	0.559	0.505	0.382	0.460	0.477	0.386	0.475

Table 3. AUC scores of nine algorithms in terms of 11 attributes

Tracker	Ours	HDT	HCF	CNN-SVM	SRDCF	Staple	LCT2	KCF	DSST
Average tracking speed	7.9	9.1	13.9	-	4.3	56.3	28.1	317	46.4

Table 4. Comparison of average tracking speeds of 9 tracking algorithmsframe/s

Chaoyi Zhang, Li Peng, Tianhao Jia, Jiwei Wen. Multi-Filter Collaborative Tracking Algorithm Based on High-Confidence Updating Strategy[J]. Laser & Optoelectronics Progress, 2019, 56(19): 191502

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information