Video analytics for metro security: pilot project at Universitat station (Barcelona)

Within the framework of the Urban Mobility Plan (PMU 2024–2030) and the Smart City Barcelona Strategy, the client faced the challenge of improving safety at metro stations

Project background


Within the framework of the Urban Mobility Plan (PMU 2024–2030) and the Smart City Barcelona Strategy, the client faced the challenge of improving safety at metro stations, reducing incident response times, and optimizing the work of security services. Manual monitoring of video streams from cameras did not allow for the timely detection of critical events: falls, fights, left-behind objects, people near the platform edge, and other threats.
 

A solution was required that could automatically detect incidents in real time, reduce the workload on operators, and ensure rapid data transmission to the control center.
 

Implementation progress


A solution for the automatic detection of security events at Universitat station was developed and implemented. The pilot project was carried out from May to September 2025.
 

Technical architecture:

IP cameras were installed at the station, and their video streams are transmitted via a Telefónica 5G router to a server with an NVIDIA T4 GPU, where the OSMOCV platform with seven neural services is deployed. Data transmission is carried out over the 5G mobile network with a latency of less than 250 ms — this allowed the solution to be deployed without laying new cable lines, using the existing mobile infrastructure.
 

Project feature: operation via 5G mobile internet

The conditions of the pilot project, defined by Telefónica, assumed the use of exclusively mobile internet for video data transmission. This imposed limitations: the traffic volume was limited to 400 GB per month, which required optimization of the system's operating schedule. Nevertheless, the solution functioned successfully within the specified time intervals, confirming the possibility of operating video analytics via mobile networks in a mode close to real time.
 

The solution included six neural services configured for operation in platform zones:

  • Human pose detection — analysis of body position: recording falls and immobility within a specified time interval (30–40 seconds);
  • Line crossing detector — recording the crossing of the yellow safety line on the platform when no train is present;
  • Left-behind object detector — detection of static objects of a specified size in the control zone;
  • Hand gesture detector — recognition of signals indicating the need for assistance;
  • Aggressive behavior detector — identification of physical conflicts between people;
  • Animal detector — detection of the presence of animals on the platform.
     

Additionally implemented:

  • face pixelation to comply with GDPR requirements (upon TMB request);
  • operator alerts via Telegram and email;
  • adaptation of neural services to real operating conditions (recalibration based on testing results).
     

Achieved results


  • High detection accuracy — the average performance across seven neural services was 94.3%, confirming the system's suitability for integration into the client's operational environment.
  • Reduced response time — the reaction time to incidents was reduced from 180 to 30 seconds (an 83% improvement).
  • Optimization of human resources — 1 operator is required for monitoring instead of 3 (a 66% reduction).
  • Reduced operational costs — by 30% compared to the current operating model.
  • Operation via mobile networks confirmed — the system functioned successfully via a 5G connection with a latency of less than 250 ms, opening up possibilities for rapid deployment at sites without developed cable infrastructure.
  • Regulatory compliance ensured — the platform complies with GDPR, Organic Law 3/2018 (LOPDGDD), and Spanish video surveillance legislation; images are processed in an anonymized form and automatically deleted within 72 hours.
     



    Negotiations are currently underway to replicate the solution at other Barcelona Metro stations. The project awaits an economic efficiency assessment and an investment decision from the client. If the results are positive, the system can be scaled to other metro lines, as well as bus stations, transfer hubs, and high-traffic urban infrastructure facilities.