Integrating Computer Vision Technologies for Smart Surveillance Purpose

Automatic detection of dangerous situations in order to ensure the safety of residents is a new step in the development of video surveillance systems in cities. And dangerous situations are often caused by deviant behavior of people: robbery, brawl, vandalism and etc. But due to the strong variability of such scenes, their detection is a challenging problem, which still remains unresolved. The key to solving this problem is the recognition of fine-grained features and events of scenes and the application of knowledge management technologies. In this paper, three computer vision technologies for detecting people, tracking people and estimating three-dimensional human poses were integrated with the aim of recognizing the actions and interactions of people in three-dimensional space. For all technologies an open source implementations were used that showed high results in popular computer vision challenges. A dataset was also created using computer graphics to test the developed system, containing scenes of the interaction of people in the city, shot under different point of views. This dataset showed that additional teaching of the human pose estimation component to handle challenging poses of people and camera viewpoints is required.