Computer vision technologies have long ceased to be the domain of laboratories. They are penetrating everyday life, and parking is no exception. At Sheremetyevo International Airport, an Intelligent Monitoring and Navigation System called "CoParking" has been implemented in the multi-level parking lots of Terminals B and C. This is not just video surveillance; it is a full-fledged computer vision system managed by an ensemble of neural networks. It addresses a range of tasks, from operational navigation to security control and integration with external digital services. In Parking C, the system has already demonstrated its effectiveness over a year of operation, while in Parking B, it is currently undergoing testing – the system is adapting to real scenarios, and engineers are fine-tuning its algorithms.
System Architecture
In the parking lots of Terminals B and C, more than 1,600 intelligent cameras are involved in total. LED occupancy indicators are installed above the parking spots, with colors changing according to the status of the spot: green – available, red – occupied. The cameras act as intelligent sensors: they not only capture images but also analyze them using deep learning technologies. Information is aggregated on a central management server, where it is processed in real-time. From the server, data is distributed in two main directions. The first is to electronic displays and interactive diagrams on the "Sheremetyevo Parking" website: this provides navigation for drivers. The second is to the CAPS, a comprehensive automated parking system. CAPS is responsible for calculating parking costs, managing access media (parking tickets, RFID tags, state registration plates – GRZ), and storing session history.
Video stream processing is based on a complex of computer vision methods. First of all, the neural network performs object detection: architectures such as YOLO (You Only Look Once) allow real-time identification of cars, people, and other significant elements in the frame. In the next step, pixel segmentation is applied to each parking spot: the system creates masks that enable high-precision determination of spot occupancy, even if the car is parked unevenly or partially obscured by another object. Moreover, analyzing these masks dynamically allows for automatic identification of parking violations: non-compliance with markings, parking a vehicle in two or more spots, stopping in driveways. State registration plates are read using Optical Character Recognition (OCR) based on neural networks – the system effectively handles dirty or poorly lit plates. Finally, algorithms are used to analyze movement and capture events (for example, accidents or suspicious activity around a car) by comparing frames and assessing distances to objects, tracking their trajectory.
After a year of operating the system in Parking Terminal C, developers, using supervised learning mechanisms, have accumulated and annotated a significant array of data from the cameras, allowing for further training of the neural networks on real scenarios. As a result, the accuracy of recognizing parking spot occupancy has reached 97–98%, and the reliability of detecting violations and incidents is confidently approaching these same figures.
This complex computer vision system is not assembled merely for the sake of technology itself; its main task is to make parking as convenient and safe as possible for the driver.
What the User Receives
For car owners, "CoParking" offers a whole set of digital services available through a unified infrastructure.
Navigation. This is the basic scenario. Upon entering the parking lot, displays show the number of available spots on each floor. At intersections, signs direct drivers to free areas. Color LED indicators light up above each block of parking spots. According to calculations, the time spent searching for a parking spot is reduced by an average of 35%.
Finding a Car. An interactive map is available on the "Sheremetyevo Parking" website, displaying real-time occupancy of spots in Parking Terminals B and C. Additionally, a function for finding a car by state number is implemented: the system shows the floor, parking spot number, and builds a route from the elevator lobby.
Security Control and Violation Detection. The system automatically identifies parking rule violations. A video fragment is generated for each event (for violations – several photo captures at specified time intervals). The access medium (ticket or GRZ) can be automatically blocked – this makes it impossible to exit the parking lot until the fine is paid. For all events, the system notifies the dispatcher and, if necessary, sends information to the website for display in the client's Personal Account.
Development Prospects
"CoParking" is an evolving platform. In the near future, additional services are planned to be implemented in the Personal Account. Among them are "Traffic Control" (the client can block their vehicle's exit via a switch in the interface; if the vehicle attempts to move, the system captures the incident, notifying the client and dispatcher) and "Online Video Surveillance" (access to a live video stream from the camera above their parking spot). These services are currently in the design stage and will be launched after completing all technical and legal procedures.
Thus, the innovation represents not just a monitoring system, but a multifunctional digital platform that transforms parking from a routine procedure into a managed service with a high level of comfort and safety.