UEFISCDI DeepVisionRomania

DeepVisionRomania “Identifying People in Video Streams using Silhouette Biometrics”, owner University “Politehnica” of Bucharest (led by AI Multimedia Lab), partners Softrust Vision Analytics, Romanian Ministry of National Defence — Military Equipment and Technologies Research Agency, public beneficiary Romanian Intelligence Service, funded by UEFISCDI, Solutions Axis, grant 28SOL/2021 (budget ~550k Eur, 2021-2023);

Strategic Relevance and Scientific Premise

This research initiative supports Romania’s strategic goals in the field of national security, video analytics, and artificial intelligence, as established in PNCDI III. The project focuses on developing an intelligent platform for the automated identification and re-identification of persons in video streams, using biometric silhouette features extracted from both static and dynamic characteristics (e.g., body proportions, gait, and attire).

The proposed system will support both operational surveillance scenarios and real-time analytics, enabling authorities to detect and track individuals in complex environments, while ensuring compliance with GDPR and data protection standards.

Project Objectives

The overarching goal is to develop, validate, and operationalize a modular software system capable of:

• Detecting and identifying persons using visual and biometric silhouette descriptors;
• Re-identifying individuals across video streams and surveillance zones;
• Incorporating selective privacy masking, including facial obfuscation and region blurring;
• Ensuring real-time video analytics on high-resolution streams (4K), with under 10s latency;
• Managing system access and audit trails via a centralized GDPR-compliant identity service.

Research Methodology and Phased Development

The research plan is structured into distinct, interconnected phases:

1. Exploration & Requirements Definition – analysis of existing technologies, identification of user needs, and definition of biometric methods and evaluation criteria;
2. Technology Development – implementation of silhouette-based recognition algorithms and architecture definition (TRL 2 to 5);
3. Module Demonstration – functional demonstrations of privacy masking and recognition components on test data (TRL 5);
4. System Integration & Validation – development of a demonstrator platform and real-time testing in operational scenarios (TRL 6–7).

Anticipated Results and Deliverables

The project will result in:

• A software suite for silhouette-based video identification, including recognition and privacy modules;
• Integration APIs and documentation for deployment in operational environments;
• Demonstrator platform enabling real-time analytics with 4K video streams;
• Encryption and masking modules with user-controlled visual obfuscation options;
• Hardware-software deployment blueprint, supporting recording from up to 12 4K video sources, and 10 TB of local storage.

The platform will be licensed for public institutional use, and relevant components will be made available under open or restricted access depending on classification level. All development will comply with national standards on the use and management of classified and personal information.

Institutional Impact and National Significance

The project reinforces our institution’s capacity in:

• Applied AI for security and surveillance;
• Privacy-aware analytics;
• Multimodal video signal processing;
• National and European innovation in biometric technologies.

Through this work, our university contributes directly to Romania’s National Strategy for Research, Development and Innovation, and strengthens its role as a provider of technological solutions to public security stakeholders.

Timeline and Project Coordination

The project is implemented over a 20-month period, from September 2021 to May 2023, funded with 550k Eur by UEFISCDI.

For Further Information

To learn more about the project, explore collaboration opportunities, or inquire about technical specifications, please contact:

📩 Prof. Bogdan IONESCU
Email: bogdan.ionescu@upb.ro