自适应特征选择的相关滤波跟踪算法

刘万军; 孙虎; 姜文涛

doi:doi:10.3788/AOS201939.0615004

光学学报, 2019, 39 (6): 0615004, 网络出版: 2019-06-17

自适应特征选择的相关滤波跟踪算法下载： 1038次

Correlation Filter Tracking Algorithm for Adaptive Feature Selection

刘万军 ¹孙虎 ^2,*姜文涛 ¹

作者单位

¹ 辽宁工程技术大学软件学院, 辽宁葫芦岛 125105

² 辽宁工程技术大学研究生院, 辽宁葫芦岛 125105

图 & 表

图 1. 整体框架示意图

Fig. 1. Schematic of overall frame

下载图片查看原文

图 2. 自适应选择检测跟踪示意图

Fig. 2. Schematic of adaptive selection detection tracking

下载图片查看原文

图 3. 重检测过程示意图

Fig. 3. Schematic of redetection process

下载图片查看原文

图 4. 目标运动轨迹图

Fig. 4. Trajectory of object motion

下载图片查看原文

图 5. 预测区域构造示意图

Fig. 5. Schematic of predicted area

下载图片查看原文

图 6. 8种跟踪算法在OTB50上的精确率和成功率。(a)精确率对比曲线;(b)成功率对比曲线

Fig. 6. Precisions and success rates for 8 tracking algorithms on OTB50. (a) Precision comparison; (b) success rate comparison

下载图片查看原文

图 7. 8种跟踪算法在OTB100上的精确率和成功率。(a)精确率对比曲线;(b)成功率对比曲线

Fig. 7. Precisions and success rates of 8 tracking algorithms on OTB100. (a) Precision comparison; (b) success rate comparison

下载图片查看原文

图 8. 8种跟踪方法在部分序列上的跟踪结果

Fig. 8. Tracking results of 8 tracking algorithms in partial sequences

下载图片查看原文

表 1融合特征多权重分配方式

Table1. Multi-weight distribution of fusion features

Number	Filter weight	Color weight	Total weight
1	0.7	0.3	1
2	0.3	0.7	1
3	0.5	0.5	1

查看原文

表 2目标中心位置比较

Table2. Comparison of central positions of targets

Frame	Detection position	Feature point	Redetection position	Actual position
1	(115.5,188.0)	-	-	(115.5,188.0)
2	(115.0,195.2)	-	-	(116.5,193.5)
3	(112.0,202.5)	-	-	(110.5,201.0)
4	(107.9,203.7)	-	-	(107.5,202.5)
5	(107.8,219.1)	178	(108.3,205.2)	(109.5,207.0)
6	(102.5,212.4)	198	(101.3,215.7)	(99.5,217.5)
7	(93.5,212.7)	226	(91.1,213.8)	(89.5,215.5)
8	(79.1,202.8)	120	(68.2,194.0)	(66.0,194.5)
9	(66.0,188.8)	-	-	(67.5,188.5)
10	(79.1,182.3)	-	-	(76.5,184.5)

查看原文

表 3估计尺度与实际尺度信息表

Table3. Estimated scale and actual scale information

Estimated scale	Frame
Estimated scale	22	32	42	52	62	72	82
1	(44,27)	(45,28)	(46,29)	(47,29)	(48,30)	(51,32)	(53,33)
2	(45,28)	(46,29)	(47,29)	(48,30)	(49,30)	(52,32)	(54,34)
3	(45,28)	(47,29)	(48,30)	(49,30)	(50,31)	(53,33)	(55,34)
4	(46,29)	(48,30)	(49,30)	(50,31)	(51,32)	(54,34)	(57,35)
5	(47,29)	(49,30)	(50,31)	(51,32)	(52,32)	(55,34)	(58,36)
6	(48,30)	(50,31)	(51,32)	(52,32)	(53,33)	(57,35)	(59,36)
7	(49,30)	(51,32)	(52,32)	(53,33)	(54,34)	(58,36)	(60,37)
8	(50,31)	(52,32)	(53,33)	(54,34)	(55,34)	(59,36)	(61,38)
9	(51,32)	(53,33)	(54,34)	(55,34)	(57,35)	(60,37)	(62,39)
10	(52,32)	(54,34)	(55,34)	(57,35)	(58,36)	(61,38)	(64,39)
11	(53,33)	(55,34)	(57,35)	(58,36)	(59,36)	(62,39)	(65,40)
12	(54,34)	(57,35)	(58,36)	(59,36)	(60,37)	(64,39)	(66,41)
13	(55,34)	(58,36)	(59,36)	(60,37)	(61,38)	(65,40)	(68,42)
14	(57,35)	(59,36)	(60,37)	(61,38)	(62,39)	(66,41)	(69,43)
15	(58,36)	(60,37)	(61,38)	(62,39)	(64,39)	(68,42)	(70,44)
16	(59,36)	(61,38)	(62,39)	(64,39)	(65,40)	(69,43)	(72,44)
17	(60,37)	(62,39)	(64,39)	(65,40)	(66,41)	(70,44)	(73,45)
18	(32,20)	(33,21)	(34,21)	(34,21)	(35,22)	(37,23)	(39,24)
19	(32,20)	(34,21)	(34,21)	(35,22)	(36,22)	(38,24)	(40,25)
20	(33,21)	(34,21)	(35,22)	(36,22)	(37,23)	(39,24)	(40,25)
21	(34,21)	(35,22)	(36,22)	(37,23)	(37,23)	(40,25)	(41,25)
22	(34,21)	(36,22)	(37,23)	(37,23)	(38,24)	(40,25)	(42,26)
23	(35,22)	(37,23)	(37,23)	(38,24)	(39,24)	(41,25)	(43,27)
24	(36,22)	(37,23)	(38,24)	(39,24)	(40,25)	(42,26)	(44,27)
25	(37,23)	(38,24)	(39,24)	(40,25)	(40,25)	(43,27)	(45,28)
26	(37,23)	(39,24)	(40,25)	(40,25)	(41,25)	(44,27)	(45,28)
27	(38,24)	(40,25)	(40,25)	(41,25)	(42,26)	(45,28)	(46,29)
28	(39,24)	(40,25)	(41,25)	(42,26)	(43,27)	(45,28)	(47,29)
29	(40,25)	(41,25)	(42,26)	(43,27)	(44,27)	(46,29)	(48,30)
30	(40,25)	(42,26)	(43,27)	(44,27)	(45,28)	(47,29)	(49,30)
31	(41,25)	(43,27)	(44,27)	(45,28)	(45,28)	(48,30)	(50,31)
32	(42,26)	(44,27)	(45,28)	(45,28)	(46,29)	(49,30)	(51,32)
33	(43,27)	(45,28)	(45,28)	(46,29)	(47,29)	(50,31)	(52,32)
Optimum scale	(41,25)	(45,28)	(46,29)	(47,29)	(49,30)	(52,32)	(53,33)
Actual scale	(43,24)	(48,25)	(46,27)	(47,27)	(53,28)	(55,26)	(54,27)

查看原文

表 48种跟踪算法平均跟踪性能比较

Table4. Average tracking performance comparison among eight tracking algorithms

Algorithm		fDSST	SAMF	SRDCF	BACF	STAPLE	STRCF	ECO	Ours
Mean DP	OTB50	0.696	0.653	0.736	0.735	0.693	0.831	0.868	0.789
	OTB100	0.742	0.741	0.778	0.801	0.795	0.855	0.886	0.819
Mean OP	OTB50	0.623	0.558	0.652	0.673	0.605	0.684	0.698	0.675
	OTB100	0.583	0.554	0.612	0.643	0.604	0.680	0.697	0.654

查看原文

表 58种跟踪算法平均跟踪速度比较

Table5. Average tracking speed comparison among 8 tracking algorithms

Algorithm		fDSST	SAMF	SRDCF	BACF	STRCF	STAPLE	ECO	Proposed
Mean FPS /(frame·s^-1)	OTB50	74.78	19.81	6.46	27.92	18.76	55.51	1.43	71.43
	OTB100	88.44	22.80	6.46	31.92	21.90	68.83	1.38	72.61

查看原文

刘万军, 孙虎, 姜文涛. 自适应特征选择的相关滤波跟踪算法[J]. 光学学报, 2019, 39(6): 0615004. Wanjun Liu, Hu Sun, Wentao Jiang. Correlation Filter Tracking Algorithm for Adaptive Feature Selection[J]. Acta Optica Sinica, 2019, 39(6): 0615004.

自适应特征选择的相关滤波跟踪算法下载： 1038次

图 1. 整体框架示意图

Fig. 1. Schematic of overall frame

图 2. 自适应选择检测跟踪示意图

Fig. 2. Schematic of adaptive selection detection tracking

图 3. 重检测过程示意图

Fig. 3. Schematic of redetection process

图 4. 目标运动轨迹图

Fig. 4. Trajectory of object motion

图 5. 预测区域构造示意图

Fig. 5. Schematic of predicted area

图 6. 8种跟踪算法在OTB50上的精确率和成功率。(a)精确率对比曲线;(b)成功率对比曲线

Fig. 6. Precisions and success rates for 8 tracking algorithms on OTB50. (a) Precision comparison; (b) success rate comparison

图 7. 8种跟踪算法在OTB100上的精确率和成功率。(a)精确率对比曲线;(b)成功率对比曲线

Fig. 7. Precisions and success rates of 8 tracking algorithms on OTB100. (a) Precision comparison; (b) success rate comparison

图 8. 8种跟踪方法在部分序列上的跟踪结果

Fig. 8. Tracking results of 8 tracking algorithms in partial sequences

表 1融合特征多权重分配方式

Table1. Multi-weight distribution of fusion features

表 2目标中心位置比较

Table2. Comparison of central positions of targets

表 3估计尺度与实际尺度信息表

Table3. Estimated scale and actual scale information

表 48种跟踪算法平均跟踪性能比较

Table4. Average tracking performance comparison among eight tracking algorithms

表 58种跟踪算法平均跟踪速度比较

Table5. Average tracking speed comparison among 8 tracking algorithms

关于本站 Cookie 的使用提示

全站搜索

自适应特征选择的相关滤波跟踪算法 下载： 1038次

图 1. 整体框架示意图

Fig. 1. Schematic of overall frame

图 2. 自适应选择检测跟踪示意图

Fig. 2. Schematic of adaptive selection detection tracking

图 3. 重检测过程示意图

Fig. 3. Schematic of redetection process

图 4. 目标运动轨迹图

Fig. 4. Trajectory of object motion

图 5. 预测区域构造示意图

Fig. 5. Schematic of predicted area

图 6. 8种跟踪算法在OTB50上的精确率和成功率。(a)精确率对比曲线;(b)成功率对比曲线

Fig. 6. Precisions and success rates for 8 tracking algorithms on OTB50. (a) Precision comparison; (b) success rate comparison

图 7. 8种跟踪算法在OTB100上的精确率和成功率。(a)精确率对比曲线;(b)成功率对比曲线

Fig. 7. Precisions and success rates of 8 tracking algorithms on OTB100. (a) Precision comparison; (b) success rate comparison

图 8. 8种跟踪方法在部分序列上的跟踪结果

Fig. 8. Tracking results of 8 tracking algorithms in partial sequences

表 1融合特征多权重分配方式

Table1. Multi-weight distribution of fusion features

表 2目标中心位置比较

Table2. Comparison of central positions of targets

表 3估计尺度与实际尺度信息表

Table3. Estimated scale and actual scale information

表 48种跟踪算法平均跟踪性能比较

Table4. Average tracking performance comparison among eight tracking algorithms

表 58种跟踪算法平均跟踪速度比较

Table5. Average tracking speed comparison among 8 tracking algorithms

相关论文

相关资讯

关于本站 Cookie 的使用提示

全站搜索

自适应特征选择的相关滤波跟踪算法下载： 1038次