140 likes | 306 Views
A Dynamic Hierarchical Clustering Method for Trajectory-Based Unusual Video Event Detection. Fan Jiang, Ying Wu , Senior Member, IEEE, and Aggelos K. Katsaggelos , Fellow, IEEE IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 18, NO. 4, APRIL 2009. Introduction. Hidden Markov Model (HMM).
E N D
A Dynamic Hierarchical Clustering Method forTrajectory-Based Unusual Video Event Detection Fan Jiang, Ying Wu, Senior Member, IEEE, and Aggelos K. Katsaggelos, Fellow, IEEE IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 18, NO. 4, APRIL 2009
Introduction • Hidden Markov Model (HMM)
Cross Likelihood Ratio (CLR)X = training trajectoryY = likelihood trajectoryλx, λy = HMM of X or Y
Bayesian Information Criterion (BIC) • Dissimilarity
TRAJECTORY CLUSTERING • Dynamic hierarchical clustering (DHC)
Fifteen categories of any three trajectory groups according to different nearest neighbors
Merging can be rejected (exclusion) if • Substituting BIC • Where :
Assume : • Sufficient condition to be satisfied
NORMAL CLUSTER IDENTIFICATION ANDABNORMALITY DETECTION • Ifthen trajectory i is unusual
Examples of normal (a)–(d) and unusual (e)–(h) trajectories.