Presentation is loading. Please wait.

Presentation is loading. Please wait.

Online Multiple Classifier Boosting for Object Tracking Tae-Kyun Kim 1 Thomas Woodley 1 Björn Stenger 2 Roberto Cipolla 1 1 Dept. of Engineering, University.

Similar presentations


Presentation on theme: "Online Multiple Classifier Boosting for Object Tracking Tae-Kyun Kim 1 Thomas Woodley 1 Björn Stenger 2 Roberto Cipolla 1 1 Dept. of Engineering, University."— Presentation transcript:

1 Online Multiple Classifier Boosting for Object Tracking Tae-Kyun Kim 1 Thomas Woodley 1 Björn Stenger 2 Roberto Cipolla 1 1 Dept. of Engineering, University of Cambridge 2 Computer Vision Group, Toshiba Research Europe

2 The Task: Object Tracking Example sequence 1 Target appearance changes due to changes in - pose - illumination - object deformation Example sequence 2

3 Learning Multi-Modal Representations - Multi-view face detection [Rowley et al. 98, Schneiderman et al. 00, Jones Viola 03] - Multi-category detection, Sharing features [Torralba et al. 04] Positive examples Negative examples

4 Joint Clustering and Training K-means clustering Face cluster 1 Face cluster 2 Positive examples Negative examples Feature pool [Kim and Cipolla 08, Babenko et al. 08]

5 Given: Set of n training samples with labels number of strong classifiers Learn strong classifiers: Combine classifier output with “Noisy OR” function Map to probabilities with sigmoid function MCBoost: Multiple Strong Classifier Boosting [Kim and Cipolla 08, Babenko et al. 08]

6 For given weights, find K weak-learners at t-th round of boosting to maximize Weak-learner weights found by a line search to maximize where Sample weight update by AnyBoost method [Mason et al. 00] MCBoost (continued)

7 MCBoost: Toy Example 1 Input dataMCBoost result (K=3)

8 Toy Example 2

9 Standard AdaBoost

10 MCBoost [Kim and Cipolla 08]

11 MC Boost with weighting function Q MCBQ

12 Classifier Assignment Make classifier assignment explicit using function weight of strong classifier on sample is updated at each round of boosting. Here: K -component GMM in d -dim eigenspace, k -th mode is area of expertise of

13 Joint Boosting and Clustering MCBoost MCBQ

14 Input: Data set, set of weak learners Output: Strong classifiers for t=1,…,T // boosting rounds for k=1,…,K // strong classifiers Find weak learners and their weights Update sample weights end MCBQ Algorithm Update sample weights Update weighting function Init with GMM Init weights to values of, weighting function

15 MCBQ for Object Tracking Principle: 1. (Short) supervised training phase 2. On-line updates

16 Online Boosting one sample Init importance Estimate errors Select best weak classifier Update weight Estimate importance Current strong classifier [Oza, Russel 01, Grabner, Bischof 06] Global classifier pool Estimate errors Select best weak classifier Update weight Estimate errors Select best weak classifier Update weight Estimate importance

17 Online MCBQ Classifiers Sample weight distribution Selector Update Selector Select weak classifiers, add to Update weights, re-normalize

18 Results

19 Improved Pose Expertise MCBoost MCBQ

20 Multi-pose Tracking with MCBQ

21 Tracking Experiments

22 Tracking “Cube” sequence MCBQMILTrackSemiBoost

23 Tracking Experiments Tracking error

24 Summary Tracking: Build appearance model, then update online No detector is required, i.e. not object specific. Handles rapid appearance changes. Simultaneous pose estimation and tracking is possible. K is currently set by hand. Incorrect adaptation may still occur. Extension of MCBoost to online setting Extension of MIL to multi-class

25 Thank you

26 Tracking: Generative vs Discriminative Generative - Eigentracking [Black, Jepson 96] - Appearance manifolds [Lee et al. 05] Discriminative - Feature selection [Collins et al. 03] - On-line boosting [Grabner et al. 06]

27 AnyBoost related Multi-component boosting [Dollar et al ECCV08] MP boosting [Babenko et al ECCVW08] MCBoost [Kim and Cipolla NIPS08]  Noisy-OR boosting for multiple instance learning [Viola et al NIPS06]

28 Tracking Experiments Tracking error. Average center location errors rounded to nearest integer (in pixels). Algorithms compared are Semi-Boost [8] (best of 5 runs), MILTrack [3], our implementations of AdaBoost, MCBoost [13] and MCBQ trackers. Bold font indicates best performance, italic second best. Cumulative errors are weighted by the number of frames per sequence.

29 Updating Weighting Function

30 Improvement by Online Updates Offline MCBQ on test set Online MCBQ on test set

31 Simultaneous Tracking and Pose Estimation side viewfront view

32


Download ppt "Online Multiple Classifier Boosting for Object Tracking Tae-Kyun Kim 1 Thomas Woodley 1 Björn Stenger 2 Roberto Cipolla 1 1 Dept. of Engineering, University."

Similar presentations


Ads by Google