Tracking humans that are interacting with the other subjects or environmentremains unsolved in visual tracking, because the visibility of the human ofinterests in videos is unknown and might vary over times. In particular, it isstill difficult for state-of-the-art human trackers to recover complete humantrajectories in crowded scenes with frequent human interactions. In this work,we consider the visibility status of a subject as a fluent variable, whosechanges are mostly attributed to the subject's interactions with thesurrounding, e.g., crossing behind another objects, entering a building, orgetting into a vehicle, etc. We introduce a Causal And-Or Graph (C-AOG) torepresent the causal-effect relations between an object's visibility fluentsand its activities, and develop a probabilistic graph model to jointly reasonthe visibility fluent change (e.g., from visible to invisible) and track humansin videos. We formulate the above joint task as an iterative search of feasiblecausal graph structure that enables fast search algorithm, e.g., dynamicprogramming method. We apply the proposed method on challenging video sequencesto evaluate its capabilities of estimating visibility fluent changes ofsubjects and tracking subjects of interests over time. Results with comparisonsdemonstrated that our method clearly outperforms the alternative trackers andcan recover complete trajectories of humans in complicated scenarios withfrequent human interactions.
展开▼
