Trajectory Analysis and Semantic Region Modeling Using Nonparametric Hierarchical Bayesian Models

摘要: We propose a novel framework of using nonparametric Bayesian model, called Dual Hierarchical Dirichlet Processes (Dual-HDP) (Wang et al. in IEEE Trans. Pattern Anal. Mach. Intell. 31:539---555, 2009), for unsupervised trajectory analysis and semantic region modeling surveillance settings. In our approach, trajectories are treated as documents observations an object on words document. Trajectories clustered into different activities. Abnormal detected samples with low likelihoods. The regions, which subsets paths commonly taken by objects related to activities the scene, also modeled. Under Dual-HDP, both number activity categories regions automatically learnt from data. this paper, we further extend Dual-HDP Dynamic model allows dynamic update models online detection normal/abnormal Experiments evaluated simulated data set two real sets, include 8,478 radar tracks collected maritime port 40,453 visual parking lot.