Prevent: Intelligent Security for Critical Infrastructure

The Challenge

In this context, PREVENT-PCP was launched as a European Pre-Commercial Procurement (PCP) project aimed at developing advanced solutions capable of automatically detecting abandoned objects, identifying their owners or responsible parties, and coordinating an effective and ethical security response.

Vicomtech was selected among the competing suppliers throughout the PCP process, achieving a high level of technological maturity (TRL 8) and ensuring that the solution could operate robustly and continuously in real, international environments.

The Proposed Solution

The response to these challenges was PREVAIL, the solution developed by the consortium led by Satways, together with imotion Analytics and Vicomtech. In its final phase, PREVAIL evolved from a laboratory prototype into an integrated pre-commercial solution, combining the CSIM security management platform with advanced AI-based video analytics.

AI and computer vision architecture

Vicomtech led the design and development of the visual intelligence algorithms, specifically optimised for high-density environments:

• Advanced detection and tracking
  Models trained to detect people (body and face) and portable objects, maintaining element identity over time and across multiple cameras.
• Owner–object association
  Algorithms that measure overlap and metric distance on the ground plane to associate objects with their owners or groups of people, significantly reducing false alarms.
• Non-biometric re-identification (Re-ID)
  A search engine based on appearance vectors (clothing, colours and shapes) that enables individuals to be located across different cameras without the need to store permanent biometric data.
• Avatar- and text-based search
  A key innovation of the project: the ability to search for suspects using textual descriptions or synthetically generated images through text-to-image techniques, streamlining the work of security operators.

In addition, to ensure reliable performance, the project addressed intensive data management:

• Large-scale image annotation
  Hundreds of thousands of real station images were annotated to improve model accuracy and robustness.
• Zero-shot models
  Use of advanced techniques to pre-annotate objects and optimise incremental model training.

The Result

In the final phase of the project, PREVAIL demonstrated its operational capability under highly demanding conditions:

• International pilots
  Deployment and evaluation of the solution in critical infrastructures in France, Spain, Portugal and Italy, each with different camera configurations and passenger flows.
• Proven scalability
  Following the initial pilots in Paris, the system was optimised to manage the simultaneous data flow from up to 300 cameras, resolving bottlenecks through software and hardware optimisations.
• Technological impact and maturity
  The project concluded with a TRL 8 solution, incorporating key improvements in the segmentation pipeline and in hybrid search mechanisms (physical appearance and authorised biometrics).

The success of PREVENT-PCP consolidates Vicomtech and its partners as European benchmarks in the development of intelligent security solutions for critical infrastructures, demonstrating that artificial intelligence can manage the complexity of public spaces in an efficient, scalable and privacy-preserving manner.