Beyond Tracking: Modelling Activity and Understanding Behaviour

摘要: In this work, we present a unified bottom-up and top-down automatic model selection based approach for modelling complex activities of multiple objects in cluttered scenes. An activity is represented on discrete scene events their behaviours are modelled by reasoning about the temporal causal correlations among different events. This significantly from majority existing techniques that centred object tracking followed trajectory matching. our approach, object-independent detected classified unsupervised clustering using Expectation-Maximisation (EM) Schwarz's Bayesian Information Criterion (BIC). Dynamic Probabilistic Networks (DPNs) formulated robust holistic scene-level behaviour interpretation. particular, developed Dynamically Multi-Linked Hidden Markov Model (DML-HMM) discovery salient dynamic interlinks processes corresponding to event classes. A DML-HMM built BIC factorisation resulting its topology being intrinsically determined underlying causality order Extensive experiments conducted captured indoor outdoor Our experimental results demonstrate performance group noisy superior compared those other comparable probabilistic networks including Multi-Observation (MOHMM), Parallel (PaHMM) Coupled (CHMM).