[论文解读] Learning Aberrance Repressed Correlation Filters for Real-Time UAV Tracking
ARCF 在 BACF 框架中抑制响应图中的异常性,从而提升无人机跟踪的鲁棒性和精度,实现实时性能,并在无人机基准测试中超越20个最先进追踪器。
Traditional framework of discriminative correlation filters (DCF) is often subject to undesired boundary effects. Several approaches to enlarge search regions have been already proposed in the past years to make up for this shortcoming. However, with excessive background information, more background noises are also introduced and the discriminative filter is prone to learn from the ambiance rather than the object. This situation, along with appearance changes of objects caused by full/partial occlusion, illumination variation, and other reasons has made it more likely to have aberrances in the detection process, which could substantially degrade the credibility of its result. Therefore, in this work, a novel approach to repress the aberrances happening during the detection process is proposed, i.e., aberrance repressed correlation filter (ARCF). By enforcing restriction to the rate of alteration in response maps generated in the detection phase, the ARCF tracker can evidently suppress aberrances and is thus more robust and accurate to track objects. Considerable experiments are conducted on different UAV datasets to perform object tracking from an aerial view, i.e., UAV123, UAVDT, and DTB70, with 243 challenging image sequences containing over 90K frames to verify the performance of the ARCF tracker and it has proven itself to have outperformed other 20 state-of-the-art trackers based on DCF and deep-based frameworks with sufficient speed for real-time applications.
研究动机与目标
- 在边界效应和外观变化(遮挡、光照、漂移)下推动鲁棒的无人机跟踪。
- 提出一种在基于 BACF 的框架中通过对响应图变化进行正则化来抑制检测过程中的异常性的方法。
- 通过裁剪矩阵和背景补丁扩大有效搜索区域,同时减小背景噪声。
- 开发高效的优化流程(在频域中的 ADMM)以在可接受的实时速度下学习对异常敏感的滤波器。
- 在 UAV123@10fps、DTB70 和 UAVDT 基准测试上,与手工特征和深度追踪器相比,展示出先进的性能。
提出的方法
- 整合裁剪矩阵以扩大与 BACF 相同的搜索区域,并将背景补丁作为负样本纳入。
- 引入一个异常惩罚项,限制连续帧之间响应图的变化(gamma 参数)。
- 将学习目标公式化为包含数据项、标准的 Tikhonov 正则化以及异常正则化项,然后变换到频域。
- 用 ADMM 求解所得的凸问题,给出 w(在空间域中的滤波器)和 g(频域表示)的闭式子问题更新。
- 在 g 更新中应用 Sherman–Morrison 优化以加速逐帧计算,从而实现实时性能。
实验结果
研究问题
- RQ1在基于 DCF 的无人机跟踪器中,是否可以在训练阶段抑制异常性以在遮挡和外观变化下提高检测稳定性?
- RQ2通过利用背景信息扩展搜索区域并抑制异常性,是否能够在无人机跟踪基准测试中提升精度和成功率?
- RQ3与 BACF 及其他最先进的跟踪器相比,异常性抑制对鲁棒性和速度有何影响?
- RQ4在使用手工特征(HOG、CN)与同时包含灰度和 CN 特征(ARCF-H 与 ARCF-HC)时,ARCF 的表现如何?
主要发现
| 跟踪器 | FPS | MSPF |
|---|---|---|
| ARCF-H | 51.2 | 19.53 |
| ARCF-HC | 15.3 | 65.36 |
| ECO-HC | 41.1 | 24.33 |
| STRCF | 22.6 | 44.25 |
| MCCT-H | 32.1 | 31.15 |
| STAPLE_CA | 37.2 | 26.88 |
| SRDCF | 11.7 | 85.47 |
| BACF | 52.5 | 19.05 |
| MUSTER | 2.1 | 476.19 |
| SAMF | 9.9 | 101.01 |
| DSST | 100.7 | 9.93 |
| KCF | 326.1 | 3.07 |
- 在 UAV123@10fps、DTB70 和 UAVDT 上,ARCF-HC 在手工特征跟踪器中在精度和 AUC 指标上表现最佳。
- ARCF-H(仅 HOG)也优于 BACF,显示出鲁棒性提升,尽管存在适度的速度权衡。
- 在跨数据集的平均表现中,ARCF 相较于 BACF 减少了与异常相关的响应图差异(例如 ARCF-HC 将响应图稳定性提高约7–8% 的报告指标)。
- ARCF 在 CPU 上保持可实现的实时速度,ARCF-HC 相对于其他手工特征跟踪器在 FPS 和 MSPF 上具有竞争力。
- 异常惩罚项在遮挡和光照变化期间有效抑制响应图的剧烈波动,从而减少跟踪漂移。
更好的研究,从现在开始
从论文设计到论文写作,大幅缩短您的研究时间。
无需绑定信用卡
本解读由 AI 生成,并经人工编辑审核。