Traffic Pattern Analysis and Anomaly Detection via Probabilistic Inference Model

摘要: In this chapter, we introduce a method for trajectory pattern analysis through the probabilistic inference model with both regional and velocity observations. By embedding Gaussian models into discrete topic framework, our uses continuous as well observations unlike existing approaches. addition, proposed framework combined Hidden Markov Model can cover temporal transition of scene state, which is useful in checking violation rule that some conflict topics (e.g., two cross traffic patterns) should not occur at same time. To achieve online learning even complexity model, suggest novel scheme instead collapsed Gibbs sampling. The two-stage greedy only efficient reducing search space but also accurate way accuracy becomes worse than batch learning. validate performance method, experiments were conducted on various datasets. Experimental results show explains satisfactorily patterns respect to understanding, anomaly detection, prediction.