自适应特征融合的多尺度核相关滤波目标跟踪 下载: 1360次
Multi-Scale Kernel Correlation Filter Algorithm for Visual Tracking Based on the Fusion of Adaptive Features
陈法领 1,2,3,4,5,*丁庆海 1,6常铮 1,2,4,5陈宏宇 1,2,3,4,5罗海波 1,2,4,5惠斌 1,2,4,5刘云鹏 1,2,4,5
1 中国科学院沈阳自动化研究所, 辽宁 沈阳 110016
2 中国科学院机器人与智能制造创新研究院, 辽宁 沈阳 110169
3 中国科学院大学, 北京 100049
4 中国科学院光电信息处理重点实验室, 辽宁 沈阳 110016
5 辽宁省图像处理与视觉计算重点实验室, 辽宁 沈阳 110016
6 航天恒星科技有限公司, 北京 100086
图 & 表
图 1. 自适应特征融合过程示意图
Fig. 1. Schematic of adaptive features fusion process
下载图片 查看原文
图 2. 权重调节系数ρ与目标跟踪性能之间的关系
Fig. 2. Relationship between weight adjustment ρ and the target tracking performance
下载图片 查看原文
图 3. 3种目标跟踪算法的距离精度曲线和重叠精度曲线。(a)距离精度;(b)重叠精度
Fig. 3. Distance precision curves and overlap precision curves of three target tracking algorithms. (a) Distance precision; (b) overlap precision
下载图片 查看原文
图 4. 所提算法对4组视频序列进行估计的目标尺度与真实的目标尺度对比。(a) Blurcar2;(b) Dog1;(c)Doll;(d) Carscale
Fig. 4. Comparisons of estimated scale by the proposed algorithm and actual scale on four sequences.(a) Blurcar2; (b) Dog1; (c) Doll; (d) Carscale
下载图片 查看原文
图 5. 不同目标跟踪算法的距离精度曲线和重叠精度曲线。(a)距离精度;(b) 重叠精度
Fig. 5. Distance precision curves and overlap precision curves of different target tracking algorithms. (a) Distance precision; (b) overlap precision
下载图片 查看原文
图 6. 5种算法对视频序列David的目标跟踪结果对比
Fig. 6. Comparison of tracking results among five algorithms on David sequence
下载图片 查看原文
图 7. 5种算法对视频序列Basketball的目标跟踪结果对比
Fig. 7. Comparison of tracking results among five algorithms on Basketball sequence
下载图片 查看原文
图 8. 5种算法对视频序列Carscale的目标跟踪结果对比
Fig. 8. Comparison of tracking results among five algorithms on Carscale sequence
下载图片 查看原文
图 9. 5种算法对视频序列Jogging1的目标跟踪结果对比
Fig. 9. Comparison of tracking results among five algorithms on Jogging1 sequence
下载图片 查看原文
图 10. 5种算法对视频序列Trellis的目标跟踪结果对比
Fig. 10. Comparison of tracking results among five algorithms on Trellis sequence
下载图片 查看原文
图 11. 5种算法对视频序列Soccer的目标跟踪结果对比
Fig. 11. Comparison of tracking results among five algorithms on Trellis sequence
下载图片 查看原文
表 1所提算法对4组存在尺度变化的视频序列的目标跟踪结果
Table1. Tracking results of the proposed algorithm on four scale variation sequences
Sequence | Mean ECL | Pd /% (ECL=20) | Po /% (So=0.5) |
---|
Blurcar2 | 3.42 | 100.0 | 100.0 | Dog1 | 3.81 | 100.0 | 100.0 | Doll | 2.26 | 99.3 | 99.6 | Carscale | 3.84 | 100.0 | 100.0 |
|
查看原文
表 2排名前10的算法对11种不同属性的评价指标Pd
Table2. Pd scores of the top ten algorithms on eleven attributes
Algorithm | IV | DEF | SV | OCC | MB | FM | IPR | OPR | OV | BC | LR |
---|
Proposed | 0.780 | 0.737 | 0.739 | 0.761 | 0.653 | 0.581 | 0.704 | 0.751 | 0.665 | 0.714 | 0.424 | DSST | 0.730 | 0.636 | 0.738 | 0.692 | 0.544 | 0.513 | 0.768 | 0.725 | 0.511 | 0.694 | 0.497 | KCF | 0.657 | 0.698 | 0.648 | 0.695 | 0.571 | 0.534 | 0.691 | 0.678 | 0.590 | 0.676 | 0.387 | Struck | 0.558 | 0.521 | 0.639 | 0.564 | 0.551 | 0.604 | 0.617 | 0.597 | 0.539 | 0.585 | 0.545 | SCM | 0.594 | 0.586 | 0.672 | 0.640 | 0.339 | 0.333 | 0.597 | 0.618 | 0.429 | 0.578 | 0.305 | CN | 0.576 | 0.607 | 0.599 | 0.621 | 0.551 | 0.482 | 0.674 | 0.645 | 0.438 | 0.629 | 0.408 | TLD | 0.537 | 0.512 | 0.606 | 0.563 | 0.518 | 0.551 | 0.584 | 0.596 | 0.576 | 0.428 | 0.349 | VTD | 0.557 | 0.501 | 0.597 | 0.545 | 0.375 | 0.352 | 0.599 | 0.620 | 0.462 | 0.571 | 0.168 | VTS | 0.573 | 0.487 | 0.582 | 0.534 | 0.375 | 0.353 | 0.579 | 0.604 | 0.455 | 0.578 | 0.187 | CXT | 0.501 | 0.422 | 0.550 | 0.491 | 0.509 | 0.515 | 0.610 | 0.574 | 0.510 | 0.443 | 0.371 |
|
查看原文
表 3排名前10的算法对11种不同属性的评价指标Po
Table3. Po of the top ten algorithms on eleven attributes
Algorithm | IV | DEF | SV | OCC | MB | FM | IPR | OPR | OV | BC | LR |
---|
Proposed | 0.712 | 0.733 | 0.721 | 0.738 | 0.591 | 0.532 | 0.674 | 0.702 | 0.672 | 0.648 | 0.419 | DSST | 0.681 | 0.610 | 0.640 | 0.632 | 0.528 | 0.503 | 0.679 | 0.632 | 0.512 | 0.627 | 0.437 | SCM | 0.568 | 0.565 | 0.635 | 0.599 | 0.339 | 0.335 | 0.560 | 0.575 | 0.449 | 0.550 | 0.308 | KCF | 0.543 | 0.628 | 0.474 | 0.580 | 0.561 | 0.523 | 0.613 | 0.579 | 0.610 | 0.630 | 0.355 | Struck | 0.491 | 0.473 | 0.471 | 0.493 | 0.518 | 0.567 | 0.528 | 0.506 | 0.550 | 0.545 | 0.410 | TLD | 0.460 | 0.456 | 0.494 | 0.468 | 0.482 | 0.473 | 0.476 | 0.497 | 0.516 | 0.388 | 0.327 | ALSA | 0.503 | 0.456 | 0.544 | 0.451 | 0.281 | 0.260 | 0.488 | 0.494 | 0.359 | 0.468 | 0.163 | CN | 0.450 | 0.511 | 0.421 | 0.479 | 0.480 | 0.437 | 0.550 | 0.501 | 0.458 | 0.531 | 0.399 | VTS | 0.503 | 0.441 | 0.453 | 0.465 | 0.328 | 0.325 | 0.477 | 0.496 | 0.508 | 0.516 | 0.183 | VTD | 0.480 | 0.443 | 0.460 | 0.468 | 0.320 | 0.319 | 0.500 | 0.510 | 0.491 | 0.515 | 0.170 |
|
查看原文
陈法领, 丁庆海, 常铮, 陈宏宇, 罗海波, 惠斌, 刘云鹏. 自适应特征融合的多尺度核相关滤波目标跟踪[J]. 光学学报, 2020, 40(3): 0315001. Faling Chen, Qinghai Ding, Zheng Chang, Hongyu Chen, Haibo Luo, Bin Hui, Yunpeng Liu. Multi-Scale Kernel Correlation Filter Algorithm for Visual Tracking Based on the Fusion of Adaptive Features[J]. Acta Optica Sinica, 2020, 40(3): 0315001.