AI Working Group

This working group is dedicated to advancing the production of high-value data in the physical security domain by adopting an object-centric data model, backed by formal semantics.

Chair: Peter Damm

Access to minutes of meetings and presentations: Repository

For older documents, see Archive.

Learn more about the AI Working Group and how to participate in this video here.

[onvifwg list=’twg_om’]| Ontology Metadata Working Group[/onvifwg]

Charter

Based on the overwhelming AI-related contributions at the Naples meetings in June 2025, the Technical Committee has decided to foster AI integration of security products by extending the scope beyond Semantic Metadata to general AI interfaces.

Scope

The workgroup works on interoperability between AI components within the safety and security industry. This will enable disparate AI-powered devices, services, and software from different manufacturers to communicate, share information, and coordinate actions in a secure and standardized manner.

Members are welcome to join forces on

  • Investigation and adaptation of existing standards: Analysing and adapting existing protocols, specifically investigating the Model Context Protocol (MCP) and the Agent-to-Agent (A2A) protocol, in relation to the unique needs of the safety and security domain.
  • Defining a common language: Establishing a standardized ontology and vocabulary for common concepts in the industry, such as “event,” “threat,” “alarm,” “location,” “person,” “vehicle,” and “action.” This will ensure that when one AI system communicates an “event,” another AI system understands what that means in a consistent way.
  • Defining interaction patterns: Specifying standard interaction patterns and workflows for multi-agent systems. This includes patterns for cooperative tasks (e.g., a camera agent and a VMS agent working together to track a suspect) and for competitive or resource-sharing scenarios.
  • Ensuring security and trust: Integrating security best practices directly into the protocol design, including secure authentication, authorization, and data integrity. This is paramount for the safety and security industry.

For the sake of clarity, a small list of what is considered out of scope:

  • Developing specific AI models or algorithms.
  • Defining the internal architecture of individual AI systems.
  • Standardizing the user interface or experience of AI-powered applications.

Dependencies

Alignment of data transport with the Video Enhancement WG