Module based on computer vision technologies using video surveillance cameras allows to automate control over an object and optimize the overall process of allowing vehicles to enter a territory with a limited access level. Various neural network architectures were selected and tested during the development process, and the optimal one was developed.

The number of images with objects that were memorized by the system is about 1 million. The system operates with both CPU and GPU. High recognition accuracy is achieved by the algorithm of developing and implementing the solution: over 70% of the training cycles are conducted in a "test environment". The current solution works using the client’s API, integration is possible in more complex conditions.