Beyond Local Search: Tracking Objects Everywhere with Instance-Specific Proposals
会议:CVPR 2016
实验室:Australian National University, Hongdong Li
目标:以前的方法只能在小范围内查找,本文的方法提供甚至在整张图上的查找的能力,实现跟踪。
特色:应用edge-based features,来自Piotr Doll´ar的一系列工作
A Super-Fast Online Face Tracking System for Video Surveillance
实验室:自动化所
目标:快速检测监控下的多个人脸,对暂时出画面的物体鲁邦。
方法:KLT + 直方图验证(保证不是背景)+ 记忆跟踪
- 直方图是做在整张脸上的。
- 开辟一个buffer用来存储跟踪消失的人脸的模型。
A Contour-Based Moving Object Detection and Tracking
2005
目标:鲁棒、快速、非刚体物体检测和跟踪
方法:edge-based features(对光照不敏感) + 梯度光流法(gradient-based optical flow technique)
Face Tracking: An implementation of the Kanade-Lucas-Tomasi Tracking algorithm
KLT在人脸跟踪上的实践
KCF [1]
- High-Speed Tracking with Kernelized Correlation Filters
- 采用判别式的tracking,需要区分目标和surrounding 环境,需要大量的训练样本,这些样本之间存在着大量的冗余,于是作者采用创新的circulant matrix来生成训练样本,这样的好处就是得到的数据矩阵是circulant,于是可以利用DFT(离散傅里叶变化)对角化,从而减少计算量
- 傅里叶变换可以把循环矩阵对角化
- 循环矩阵是一种特殊形式的 Toeplitz矩阵,它的行向量的每个元素都是前一个行向量各元素依次右移一个位置得到的结果。由于可以用离散傅立叶变换快速解循环矩阵,所以在数值分析中有重要的应用。
MOSSE[2]
- Matlab上,对640*480的图片不能实时
- 但是文章称在Python using the PyVision library,OpenCV, and SciPy上可以达到669的帧率
- 通过仿射变换得到一系列的训练数据f和g,计算所需要的模板h。在下一帧,同一个框内,计算得到最高的响应位置就是新的框中心。
“Learning to Track at 100 FPS with Deep Regression Networks”
- http://davheld.github.io/GOTURN/GOTURN.html
- 速度最快的神经网络跟踪算法
- ECCV 2016
- 但是在CPU上的速度仅有2.7fps,不能容忍
MDnet[3],
- Learning Multi-Domain Convolutional Neural Networks for Visual Tracking
- VOT 2015冠军
- http://cvlab.postech.ac.kr/research/mdnet/
DSST[4],
- Danelljan等人对基于CF的方法作了改进,增加了对缩放的估计。
- 声称快速并且高效
- 缩放方法用在MOSSE跟踪方法上,但是该缩放方法可以普遍用于其他跟踪方法
- 其fast scale search速度为:24 fps
- 提出的Exhaustive Scale Space Tracking就是将原来二维图像的通过金字塔弄成三维的,h和g也相应变成三维的。响应最大的那个层就是scale的最佳值,0.96FPS
LCT[5]
Visual Tracking: An Experimental Survey [6]
- 主要贡献:systematic analysis and the experimental evaluation of online trackers
- 在130段视频上进行评测
- 不评价off-line的算法
- 不评价contour-based算法,因为初始化比较困难
- 表1,总结了各种评价标准
- F score:
一些数据库
- OTB50 http://cvlab.hanyang.ac.kr/tracker_benchmark/datasets.html
- OTB100
- VOT2014[7] http://www.votchallenge.net/
- VOT2015
- BoBoT dataset:“D. A. Klein, D. Schulz, S. Frintrop, and A. B. Cremers, “Adaptive
- real-time video-tracking for arbitrary objects,” in Proc. IEEE IROS, Taipei, Taiwan, 2010, pp. 772–777.”
- CAVIAR dataset:few but long and difficult videos
- i-LIDS Multiple-Camera Tracking Scenario
- 3DPeS dataset:contains videos with more than 200 people walking as recorded from eight different cameras in very long video sequences
- PETS-series:
- TRECVid video dataset:large video benchmark
- ALOV++ dataset:proposed by [6]; more than 300 video sequences; http://crcv.ucf.edu/data/ALOV++/
评价标准
- PETS:Performance Evaluation of Tracking and Surveillance
- PETS and VACE,CLEAR:for evaluating the performance of multiple target detection and tracking
参考文献
[1] J. F. Henriques, R. Caseiro, P. Martins, and J. Batista, “High-Speed Tracking with Kernelized Correlation Filters,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 37, pp. 583-596, 2015.
[2] D. S. Bolme, J. R. Beveridge, B. A. Draper, and Y. M. Lui, “Visual object tracking using adaptive correlation filters,” in 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2010, pp. 2544-2550.
[3] H. Nam and B. Han. (2015, October 1, 2015). Learning Multi-Domain Convolutional Neural Networks for Visual Tracking. ArXiv e-prints 1510. Available: http://adsabs.harvard.edu/abs/2015arXiv151007945N
[4] M. Danelljan, G. Häger, F. Khan, and M. Felsberg, “Accurate scale estimation for robust visual tracking,” in British Machine Vision Conference, Nottingham, September 1-5, 2014, 2014.
[5] C. Ma, X. Yang, Z. Chongyang, and M. H. Yang, “Long-term correlation tracking,” in 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, pp. 5388-5396.
[6] A. W. M. Smeulders, D. M. Chu, R. Cucchiara, S. Calderara, A. Dehghan, and M. Shah, “Visual Tracking: An Experimental Survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 36, pp. 1442-1468, 2014.
[7] M. Kristan, J. Matas, A. Leonardis, T. Vojíř, R. Pflugfelder, G. Fernández, G. Nebehay, F. Porikli, and L. Čehovin, “A Novel Performance Evaluation Methodology for Single-Target Trackers,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 38, pp. 2137-2155, 2016.