Robust Object Tracking with Online Multiple Instance Learning Boris Babenko, Ming-Hsuan Yang, Serge Belongie. Robust Object Tracking with Online Multiple Instance Learning. IEEE Trans. on PAMI , 2011. Advisor: Sheng-Jyh Wang Student: Pei Chu
Outline Introduction Tracking by Detection(Related Work) Multiple Instance Learning (MIL) Online MILboost Experiments Conclusion
Introduction: Tracking Problem: track arbitrary object in video given location in first frame Typical Tracking System: Appearance Model Color , subspaces, feature,etc Optimization/Search Greedy local search, etc [Ross et al. ‘07]
Tracking by Detection Recent tracking work Focus on appearance model Borrow techniques from object detection Slide a discriminative classifier around image [Collins et al. ‘05, Grabner et al. ’06, Ross et al. ‘08]
Tracking by Detection: Online AdaBoost Grab one positive patch, and some negative patch, and train/update the model. negative positive Classifier Online classifier (i.e. Online AdaBoost)
Tracking by Detection Find max response negative positive old location new location X X Classifier Classifier
Tracking by Detection Repeat… negative negative positive positive Classifier Classifier
Problems What if classifier is a bit off? Tracker starts to drift How to choose training examples?
Multiple Instance Learning (MIL) Instead of instance, get bag of instances Bag is positive if one or more of it’s members is positive Positive Negative [Keeler ‘90, Dietterich et al. ‘97] [Viola et al. ‘05]
Multiple Instance Learning (MIL) MIL Training Input The bag labels are defined as:
Update all M classifiers Online MILBoost Frame t Frame t+1 Get data (bags) Update all M classifiers in pool Greedily add best K to strong classifier
Boosting Train classifier of the form: where is a weak classifier Can make binary predictions using [Freund et al. ‘97]
Online MILBoost At t frame, Update all M candidate classifiers Pick best K in a greedy fashion (M>>K) [Grabner et al. ‘06]
Online MILBoost Objective to maximize: Log likelihood of bags: where: Noisy-OR Model, The bag probability The instance probability [Viola et al. ’05, Friedman et al. ‘00]
Online MILBoost(OMB) M>K, M :is total weak classifier candidates K : is choosing the best K classifiers
Online MILBoost VS Online Adaboost
System Overview: MILtrack
Experiments Compare MILTrack to: OAB1 = Online AdaBoost w/ 1 pos. per frame OAB5 = Online AdaBoost w/ 45 pos. per frame SemiBoost = Online Semi-supervised Boosting FragTrack = Static appearance model [Grabner ‘06, Adam ‘06, Grabner ’08]
Results
Results
Results Best Second Best
Conclusions Proposed Online MILBoost algorithm Using MIL to train an appearance model results in more robust tracking