Entity appearance model generation for multimedia events in surveillance videos

Traditionally, surveillance systems only focus on a small number of entities (such as humans, entrance and exit areas, etc.) and appearance models of these entities are uploaded manually in the system. However, as the end users are becoming more aware of the vision based technologies, there is ever growing demand for advanced surveillance systems which can detect complex abnormal events on different aspects of operational activities and can also provide intelligence to improve their operational management process. To achieve this goal, we proposed the event mining framework which explores the relationship between entity feature-sets and associated text strings to generate appearance models of all the entities automatically and can update them dynamically.