這篇論文提出了一種使用多種特徵做物體追蹤的演算法，其中包含了行為預測模組(particle filter)和漸進式分類模組(Fisher linear discriminant model(FLD))。為了解決偏移問題，漸進式分類模組的建立考慮背景的資訊，以將追蹤目標和背景分得更開。並提出一種估計每個特徵的可靠性的方法，用來計算行為預測模組的觀測機率值。此外，為了解決目標物體的外觀變化和背景的變化，我們用協同式訓練的方法(co-training)更新每個特徵的漸進式分類模組。實驗結果證明，此演算法能夠處理目標的外觀變化，包括旋轉，局部遮庇，光線變化，和遠近或視點的變化。

This paper presents a multi-feature integrated algorithm which incorporates the particle filter and Fisher linear discriminant (FLD) model for object tracking. To solve the drift problem, the discriminant model for each feature is built up by considering background information to separate the object from background clutter. Similar to Adaboost method, each feature’s reliability is determined by the proposed measuring method, which is used for successive calculation of observation probability in a particle filter. Moreover, to address the object appearance variations and background changes, the FLD model for each feature is renewed by data which is selected according to the co-training concept. Experimental results showed the proposed multi-feature integrated algorithm is able to handle object appearance variations including out-of-plane rotation, partial occlusions, varying illuminations, and scale or viewpoint changes.

[1] M. S. Arulampalam, S. Maskell, N. Gordon, and T. Clapp. A tutorial on particle filters for online nonlinear/non-gaussian Bayesian tracking. IEEE Trans. on Signal Processing, 50(2):174-188, 2002.

[2] S. Avidan. Ensemble tracking. IEEE Trans. on PAMI, 29(2): 261-271, 2007.

[3] M.J. Black and A. Jepson. Eigentracking: robust matching and tracking of articulated objects using a view-based representation. ECCV, 329-342, 1996.
[4] Y. Freund. Boosting a weak learning algorithm by majority. COLT, 202-216, 1990.

[5] H. Grabner, M. Grabner, and H. Bischof. Real-time tracking via on-line boosting. British Machine Vision Association, 2006.

[6] M. Isard and A. Blake. Condensation- conditional density propagation for visual tracking. IJCV, 29(1):5-28, 1998.

[7] A. Jepson, D. Fleet, and T. El-Maraghi. Robust online appearance models for visual tracking. IEEE Trans. on PAMI, 25(10):1296-1311, 2003. (WSL)

[8] R.S. Lin, M.H. Yang and S.E. Levinson. Object tracking using incremental Fisher discriminant analysis. International Conf. on Pattern Recognition, 2:757-760, 2004.

[9] E. Maggio, F. Smerladi, and A. Cavallaro. Adaptive multifeature tracking in a particle filtering framework. IEEE Trans. on CSVT, 17(10):1348-1359, 2007.

[10] F. Tang, S. Brennan, Q. Zhao, and H.Tao. Co-tracking using semi-supervised support vector machines. International Conf. on Computer Vision. 1-8, 2007.

[11] http://www.ces.clemson.edu/~stb/research/headtracker/seq

[12] http://seqam.rutgers.edu/projects/motion/face/face.html