Multi-Scale Context-Aware Correlation Filter Tracking Algorithm Based on Channel Reliability

Mingfeng Yin; Yuming Bo; Jianliang Zhu; Panlong Wu

doi:10.3788/AOS201939.0515002

Journals >Acta Optica Sinica >Volume 39 >Issue 5 >Page 0515002 > Article

Acta Optica Sinica
Vol. 39, Issue 5, 0515002 (2019)

Multi-Scale Context-Aware Correlation Filter Tracking Algorithm Based on Channel Reliability

Mingfeng Yin^*, Yuming Bo, Jianliang Zhu, and Panlong Wu

Author Affiliations

School of Automation, Nanjing University of Science & Technology, Nanjing, Jiangsu 210094, China

show less

DOI: 10.3788/AOS201939.0515002 Cite this Article Set citation alerts

Mingfeng Yin, Yuming Bo, Jianliang Zhu, Panlong Wu. Multi-Scale Context-Aware Correlation Filter Tracking Algorithm Based on Channel Reliability[J]. Acta Optica Sinica, 2019, 39(5): 0515002 Copy Citation Text

show less

Fig. 1. Framework of proposed algorithm

Download full size

Fig. 2. Illustration of channel reliability. (a) Response map of 6th frame; (b) 6th frame; (c) 108th frame; (d) response map of 108th frame

Download full size

Fig. 3. Curves of distance precision and success rate of eight algorithms. (a) Distance precision curves of overall; (b) Success rate curves of overall

Download full size

Fig. 4. Distance precision and success rate curves for different attribute scenes. (a) Illumination variation; (b) scale variation; (c) occlusion; (d) rotation variation

Download full size

Fig. 5. Tracking results using different approaches. (a) Lemming; (b) Liquor; (c) Shaking; (d) Singer2; (e) Soccer; (f) Terris

Download full size

No.	Tracker	Feature	Scale adaption	Context-aware	Channel reliability
1	CSK	Raw pixel	○	○	○
2	KCF_MTSA	HOG	√	○	○
3	SAMF	Raw pixel+HOG+CN	√	○	○
4	MOSSE_CA	Raw pixel	○	√	○
5	DCF_CA	HOG	○	√	○
6	SAMF_CA	Raw pixel+HOG+CN	√	√	○
7	RPT	HOG	√	○	○
8	Proposed method	Raw pixel+HOG+CN	√	√	√
Note: √ means yes, ○ means no.

Table 1. Characteristics of testing tracking algorithms

Sequence	Image size /(pixel×pixel)	Target size /(pixel×pixel)	Characteristic
Lemming	640×480	61×103	IV, SV, OCC, FM, OPR, OV
Liquor	640×480	73×210	IV, SV, OCC, MB, FM, OPR, OV, BC
Shaking	624×352	61×71	IV, SV, IPR, OPR, BC
Singer2	624×352	67×122	IV, DEF, IPR, OPR, BC
Soccer	640×360	67×81	IV, SV, OCC, MB, FM, IPR, OPR, BC
Terris	320×240	68×101	IV, SV, IPR, OPR, BC

Table 2. Characteristics of image sequences in the experiment

Sequence	Proposed	CSK	MOSSE_CA	DCF_CA	SAMF	KCF_MTSA	SAMF_CA	RPT
Lemming	9.048	114.687	95.529	152.994	15.136	13.452	11.258	42.699
Liquor	7.409	158.997	95.547	19.996	10.578	8.641	9.253	38.641
Shaking	8.651	14.813	73.448	90.76	44.502	48.466	10.628	8.978
Singer2	9.919	185.826	176.144	11.402	13.238	11.124	13.256	10.837
Soccer	19.502	63.940	41.057	23.758	17.304	100.238	28.729	10.826
Terris	4.309	26.533	72.579	7.635	11.661	13.081	7.421	4.881

Table 3. Comparison of average CLEpixel

Sequence	Proposed	CSK	MOSSE_CA	DCF_CA	SAMF	KCF_MTSA	SAMF_CA	RPT
Lemming	1.000	0.462	0.540	0.299	0.739	0.842	0.915	0.635
Liquor	1.000	0.289	0.211	0.871	0.839	0.879	0.895	0.871
Shaking	1.000	0.523	0.222	0.802	0.745	0.723	0.904	0.921
Singer2	1.000	0.063	0.060	0.753	0.826	0.912	0.902	0.926
Soccer	0.931	0.263	0.556	0.895	0.895	0.247	0.881	0.989
Terris	1.000	0.612	0.681	0.973	0.918	1.000	0.856	1.000

Table 4. OP of different algorithms

Sequence	Proposed	MOSSE_CA	CSK	DCF_CA	KCF_MTSA	SAMF	SAMF_CA	RPT
Lemming	12.156	89.766	181.341	53.520	22.165	14.091	13.756	1.036
Liquor	19.256	203.726	454.252	89.831	43.680	23.492	20.101	2.258
Shaking	14.941	121.417	234.255	66.328	27.799	17.272	16.445	1.267
Singer2	9.668	80.358	154.209	25.617	16.076	12.856	11.360	0.895
Soccer	25.253	115.114	289.206	60.062	29.231	31.097	26.772	3.653
Terris	16.964	84.671	172.258	48.411	23.354	20.742	18.013	1.518

Table 5. Tracking speeds of different trackersframe·s-1

Mingfeng Yin, Yuming Bo, Jianliang Zhu, Panlong Wu. Multi-Scale Context-Aware Correlation Filter Tracking Algorithm Based on Channel Reliability[J]. Acta Optica Sinica, 2019, 39(5): 0515002

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information