Object Tracking Paper(5)--Understanding and Diagnosing Visual Tracking Systems

本文剖析了视觉跟踪系统,将其分解为运动模型、特征提取器、观测模型、模型更新器及集合后处理器五个部分。研究发现特征提取器对于跟踪效果至关重要,而强大的特征如HOG特征下,SVM和支持向量回归之间的差异很小。模型更新策略显著影响跟踪性能,过频更新会导致漂移问题。有效的集合后处理能提升整体表现。

摘要生成于 C知道 ,由 DeepSeek-R1 满血版支持, 前往体验 >

The fifth Paper: Understanding and Diagnosing Visual Tracking
Systems/ Author: Naiyan Wang, Jianping Shi, Dit-Yan Yeung, Jiaya Jia/
Publication information: ICCV 2015
Outline: This paper about tracking aims to break the DCF trackers into
five constituent components, namely, motion model, feature extractor,
observation model, model updater, ensemble post-processor. They control
the variation of one part, and conduct the experiment to find the key
factor in visual object tracking. They conclude that the feature extractor
plays the most important part in tracking. And when the features is
powerful enough, such as HOG, the difference between SVM and ridge
regression is very small. The model updating scheme can impact the
tracker’s performance significantly. The Excessive updating frequency
may cause drift problem and let the model contaminated by bad samples.
And an effective ensemble processor can improve the performance. The
diversity of the trackers could promote the robustness and accuracy.
Advantages: This conclusion is reliable and reasonable. They tell the
researchers what should be put much effort on.
Disadvantages: The features they discuss is hand-crafted and deep
features are not included. The trackers speed should be considered.
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值