Phase 1: Study the available AI equipment, techniques, and architectures. Subsequently, limitations and constraints were identified, along with opportunities for improving the technology to be adopted in alignment with the specific requirements.
D1.1. This activity aimed to define the necessary requirements for designing a system dedicated to monitoring and managing road traffic, based on advanced artificial intelligence technologies. The operational and functional requirements of the beneficiary were defined, forming the foundation for the development of the system architecture and usage scenarios. A modular and scalable system design was proposed, featuring key functionalities such as vehicle identification using LPR technology, recognition of vehicle type, model, and brand, driver and passenger detection, aggressive behavior analysis, and data encryption to ensure information security. Each component will undergo individual and integrated testing under real usage conditions to validate compliance with technical and operational requirements. The process will also include training for the beneficiary’s personnel to ensure optimal system usage and the development of comprehensive technical documentation, including detailed specifications and user guides. The system will be evaluated based on technical performance indicators, confirming the positive impact of the solution. The activity concluded with the design of the prototype, its validation, and preparation for implementation, marking a critical milestone in delivering an efficient solution aligned with the beneficiary’s needs.
D1.2. The activity focused on the evaluation and analysis of advanced algorithms used for vehicle identification and monitoring, driver behavior detection, and trajectory prediction, all aimed at improving traffic safety and efficiency. The study includes a detailed analysis of both traditional methods and deep learning-based approaches, highlighting the advantages and limitations of each. The activity involved a comprehensive review of the specialized literature, emphasizing visual detection algorithms, OCR technologies for license plate recognition, natural language processing technologies, and biometric solutions for driver identification. Additionally, the performance of these methods was analyzed in real-world contexts, along with their limitations related to the complexity of scenarios and the computational resources required.
D1.3. The primary objective of the activity was to identify the technical solutions necessary for developing an advanced system for the identification and analysis of vehicles and their occupants, in accordance with the beneficiary’s requirements. During this activity, technical requirements were analyzed, existing solutions were evaluated, and new approaches were designed to address the identified challenges. Proposed solutions included vehicle detection and analysis based on visual characteristics, image preprocessing through noise reduction and rectification and license plate recognition. Additionally, methodologies were developed for analyzing vehicle occupants, including driver identification and estimation of relevant characteristics. The activity demonstrated the feasibility of the proposed solutions, which are to be implemented in future stages of the project. The results provide a solid foundation for developing high-performance, innovative, and adaptable systems that meet current and future requirements in traffic monitoring and road safety.
D1.4. The activity pursued two main objectives: conducting a detailed analysis of data sensors to identify suitable sources for the system and presenting the necessary standards and protocols to ensure compatibility and subsequent integration into the final platform. The sensor analysis led to the identification of two main categories to be integrated into the system: video data sensors and traffic sensors. For video sources, the system will allow the integration of both IP cameras and analog cameras. IP cameras can be either standard models or advanced devices, such as LPR cameras with embedded processing. For analog cameras, integration into the platform will be achieved by using encoding systems that will convert the analog signal into an IP-compatible format. Regarding traffic sensors, the solution will be compatible with a wide range of devices, including radar sensors and inductive sensors, ensuring diverse data collection. To guarantee connectivity and interoperability between the platform and data sources, regardless of the manufacturer, established standards and protocols will be employed. The ONVIF standard is central to this process due to its role in facilitating compatibility between IP security devices from different manufacturers. It defines a set of standardized interfaces that allow uniform integration of data sources. Additionally, other relevant protocols, such as GigE, RTP, and RTSP, were analyzed for video data transmission, as well as HLS and DASH for secure on-demand distribution of multimedia content.
D1.5. The activity focused on the analysis and implementation of internal policies and procedures for the protection of personal data, in the context of European and national legal regulations, with an emphasis on the General Data Protection Regulation (GDPR). The main principles of data processing, such as transparency, data minimization, and security, were presented, highlighting their importance in organizational activities.Technical and institutional measures necessary for data protection were discussed, including encryption, backups, access control, and continuous system monitoring. The importance of training project team members and fostering an institutional culture aware of cybersecurity risks was emphasized, with periodic security testing being recommended. Policies regarding data collection, storage, and handling were addressed, including procedures for data transfers and managing security incidents, ensuring organizations can respond quickly to potential breaches. Finally, the need for periodic review of internal policies to align with new regulations and emerging threats was underlined, promoting an effective and compliant data protection framework. Proper implementation of these measures ensures the protection of personal data, compliance with regulations, and increased trust in organizations managing such data.
D1.6. The activity aimed to collect and organize public annotated datasets necessary for the development, testing, and validation of technological solutions within the project. Initially, relevant scenarios for the use of annotated data were identified, including vehicle detection, license plate recognition using OCR, vehicle classification by type, model, and brand, driver identification based on biometric features, and vehicle trajectory prediction in traffic. Based on these scenarios, relevant public datasets were gathered and documented, with each dataset analyzed in detail to highlight its source, structure, available annotations, and potential applications. These datasets were organized into categories corresponding to the identified scenarios, ensuring easy and efficient data utilization. Special attention was given to compliance with privacy and data usage regulations, with a focus on meeting GDPR requirements for datasets containing biometric information. The main outcomes of this activity were the creation of a centralized and well-documented resource of datasets and the facilitation of project team access to these resources. This resource is a key element for implementing technological scenarios and achieving the overall objectives of the project.
D1.7. The activities undertaken focused on the design, development, and integration of a complex software system aimed at meeting the requirements for advanced data processing, trajectory prediction, user access control, and intelligent report generation. Modular design was essential, with each component tailored to address specific needs while integrating seamlessly into the final system. A key aspect of the process was the use of modern technologies such as Docker and Kubernetes, enabling efficient containerization and orchestration of software components, with an emphasis on security, scalability, and performance. The integration of GIS maps, along with vector and raster layers, was crucial for accurately modeling trajectories, significantly enhancing system functionality. In terms of access control, the implementation of Active Directory and the use of the LDAP protocol provided a robust and secure mechanism for managing users and permissions. The reporting module was developed to allow for detailed, real-time reports, featuring advanced filtering options and leveraging the PostgreSQL database. The system diagram and workflow were constructed to ensure a clear and efficient process for data acquisition, processing, and utilization, culminating in comprehensive operational auditing.
Phase II: Design, development, and implementation of key AI algorithms and software modules, including vehicle and driver identification, behavior analysis, trajectory prediction, and data encryption
Requirements and architecture were refined, technological constraints and opportunities were assessed, and a scalable real-time video management architecture was defined. The Video Management Software platform was adapted, datasets were created, and prototype modules were tested and validated, laying the groundwork for full system integration and demonstration. The following deliverables have been completed:
D.2.1. During this phase, the project advanced the development of an integrated AI system for automatic vehicle and driver analysis. Requirements and high-level specifications were refined, establishing a scalable and secure architecture for visual perception, semantic interpretation, video-stream processing, and data management. Innovative algorithms were developed for multimodal vehicle identification using OCR and natural-language models, as well as for recognizing visual attributes such as brand, model, and vehicle geometry. Additional components addressed the detection of aggressive driving behaviors and the prediction of future trajectories through spatio-temporal analysis and statistical lane estimation. The system was further enhanced with biometric modules capable of detecting and identifying drivers using face- and body-based representations, robust even in challenging visual conditions. All modules were implemented, validated on real data, and integrated into the operational workflow, providing a coherent foundation for the subsequent stages of system integration and deployment.
D.2.2. This activity defined the security requirements for the ALPR system and designed a tailored cryptographic module. After assessing data risks and evaluating multiple cryptographic approaches, a complete key-management mechanism and the encryption points within the ALPR workflow were established. A functional prototype was developed and tested, confirming proper integration of encryption and compliance with all security specifications.
D.2.3. This activity delivered a scalable real-time video-management architecture. The team analyzed multiple types of video sources and designed a flexible module that supports easy integration, configuration, and removal of cameras. Standardized protocols and Docker-based containerization ensured unified processing and both vertical and horizontal scalability. The resulting platform is stable, extensible, and fully aligned with the operational requirements for real-time video processing.
D.2.4. This activity adapted the Video Management Software platform to the project’s needs by designing a user-friendly interface and implementing a secure user-management system. The result is a functional, adaptable interface paired with a robust access-control system that meets all operational and security requirements.
D.2.5. This activity created and organized the datasets needed for training, validating, and testing all AI modules, covering tasks such as vehicle identification, license-plate recognition, make/model classification, driver detection, and aggressive-behavior analysis. Using these datasets, all prototype modules were subsequently tested and validated, confirming accuracy, stability, and alignment with project requirements.
D.2.6. This activity validated the integrated software platform and its key modules—data processing, GIS mapping, and intelligent reporting—ensuring real-time performance and compliance with operational requirements. Tests confirmed reliable containerized processing via Docker, interactive GIS-based camera management, and effective real-time and archival reporting. The platform was shown to be fully functional, scalable, and well aligned with the project’s objectives.
D.2.7. This activity produced the full testing procedures for the integrated system, covering all AI modules and operational scenarios. Using public datasets and realistic internal tests, the team evaluated the system both modularly and end-to-end, confirming real-time performance and stability while identifying areas for future optimization.
D.2.8. This activity integrated the hardware–software prototype into a unified TRL6 system, combining all algorithmic modules, the software platform, and the hardware infrastructure into a functional real-time solution. Authentication, access control, video-source management, processing workflows, event visualization, reporting, and GIS-based mapping were validated, demonstrating full interoperability and stable operation.
D.2.9. This activity validated the integrated solution in realistic operating conditions, confirming correct handling of video sources, reliable execution of processing algorithms, and effective real-time and archival reporting. All critical platform functions were tested, demonstrating stable operation and accuracy. The solution was successfully validated and deemed compliant with the technical requirements for internal approval.