Software Solution for Industrial Video Analytics Systems Prototyping

Methods of solving production video analytics tasks have a wide variability, which is dictated by the high rate of computer vision technology development and the emergence of a large number of new application solutions. As a result, the process of designing systems for the peculiarities of a particular task becomes more complicated. The paper describes a software solution developed for prototyping intelligent video analytics systems, implemented in Python language. The solution provides a pipline that implements the basic functionality of video analytics - detection and tracking of objects simultaneously by several video sources, logging of events. The proposed solution has the following implementation features: neural network module of object detection, suitable for application in scenes with significantly changing scale of observed objects; possibility of simple functionality extension by adding new processing modules; built-in automatic saving of system startup parameters in the database to ensure repeatability of experiments.