A System of Objective Control at Facilities of any Purpose in Real Time

In all types of premises, whether industrial, administrative or residential, it is important to monitor various emergency situations that may pose a physical threat to people present. These threats can be caused by unlawful or unauthorised actions of other people, failure to comply with safety procedures on construction sites, or the presence of people in restricted areas. It is particularly important to identify events that pose a threat to people's lives or health and to take immediate action to eliminate the threat. The aim of this diploma thesis is to develop a video analytics system that combines high accuracy in event detection with high processing speed for searching through video streams. This goal imposes constraints on the architecture of the neural networks used in the system and presents challenges in the development of the architecture of the system itself. In order to find the best solutions to achieve the stated goal, a series of tests are performed to design the system architecture and select neural network architectures based on the existing constraints. As a result of this thesis, a video analytics service has been developed to improve the security level of the facility. It includes the necessary set of neural networks and decision support microservices to analyse and search for different emergency events captured by external surveillance cameras. These events are related to unlawful actions, violations, potentially dangerous situations, and immediate notifications are generated to the responsible units so that they can take the necessary actions.