Use of Multiple Distributed Process Instances for Activity Analysis in Videos

Video surveillance of security-relevant areas is being used ever more frequently. Because of limited human resources, they are usually only checked for the presence of problematic activity after a specific event has occurred. An approach to the solution is provided by automated systems that are capable of detecting and analyzing movement sequences of objects including persons. Even though solutions already exist for scene recognition [3, 4, 8, 9], their architecture, their problem-specific domain and the nature of the systems make it difficult to integrate new activities or better algorithms. A system structure based on decentralized process instances and their communication via defined interfaces would instead enable a simpler expansion of the system. This paper describes the determination of activities in videos using decentralized frameworks and their interconnection via standardized interfaces. All components act autonomously and provide their data via a central location. Based on this approach, the modular system can be used for a wide variety of applications in the context of machine learning.