Event classification for automatic visual-based surveillance of parking lots

In this paper, a visual-based surveillance system for real-time event detection and classification in parking lots is presented. The focus is on the high-level part of the system, i.e., the event recognition (ER) module, which is able to analyze two kinds of events (i.e., simple and composite events) that occur in the observed scene. Simple events are represented by single moving objects, e.g., vehicles, pedestrians, etc., while a composite event is represented by a set of temporally consecutive simple events, e.g., people exiting a car just entered in the parking area. An adaptive high order neural tree (AHNT) is applied for recognizing both objects and complex events.