Human activity monitoring by local and global finite state machines

There are a number of solutions to automate the monotonous task of looking at a monitor to find suspicious behaviors in video surveillance scenarios. Detecting strange objects and intruders, or tracking people and objects, is essential for surveillance and safety in crowded environments. The present work deals with the idea of jointly modeling simple and complex behaviors to report local and global human activities in natural scenes. Modeling human activities with state machines is still common in our days and is the approach offered in this paper. We incorporate knowledge about the problem domain into an expected structure of the activity model. Motion-based image features are linked explicitly to a symbolic notion of hierarchical activity through several layers of more abstract activity descriptions. Atomic actions are detected at a low level and fed to hand-crafted grammars to detect activity patterns of interest. Also, we work with shape and trajectory to indicate the events related to moving objects. In order to validate our proposal we have performed several tests with some CAVIAR test cases.

► Simple and complex behaviors are jointly modeled to report local and global human activities in natural scenes.
► Human activities are modeled with state machines.
► Knowledge about the problem domain is incorporated into an expected structure of the activity model.
► Atomic actions are detected at a low level and fed to hand-crafted grammars to detect activity patterns of interest.