The IP-Broadcast research project investigates a high-performance, low-latency IP live production architecture as an advancement of traditional broadcast workflows. The focus is on a live event scenario that demands minimal delay and seamless synchronization of multiple media sources.

A fully IP-based approach was chosen for integration into existing systems. The goal is an IP-based production workflow extended through AI-driven metadata detection. Video, audio, and metadata streams are transmitted separately yet precisely synchronized using packet-based architectures built on IP standards. Precision Time Protocol (PTP) enables accurate time-level synchronization across all media streams.

The ongoing development takes place in a controlled, reproducible test environment. At its core lies a custom-developed, web-based playout system serving as a platform for connecting and testing proprietary AI-supported analysis modules. A key component of the ongoing research is a newly designed generic metadata model for the structured description of event-based information using AI. This model allows for the distinct classification of event types (e.g., motion, object interaction, positional change) with exact time references and hierarchical structuring. Metadata is intended to be embedded as separate IP-broadcast–compliant streams using standardized formats such as JSON or XML. The model is designed to support automatic event recognition based on previously annotated training data and aims to establish a scalable metadata processing foundation for real-time applications.

The modeling and intended implementation follow a modular architecture approach that separates raw data acquisition, semantic interpretation, and synchronized playback. This creates a continuous link between live video and event interpretation to support time-critical production processes through automation.

As part of the ongoing project, the developed infrastructure comprises a functional playout server, an AI framework for event-based analysis, and successfully conducted tests of time-synchronized multicast transmission. These components form the basis for a scalable, modular live production architecture. In the future, visually lossless low-delay codecs such as JPEG XS will be evaluated with respect to processing speed, image quality, and system integration. Additionally, real-time optimization of AI processing on CPU/GPU systems is planned.