Automating Live Streaming Ad Operations with AI-Based SCTE-35 Auto-Insertion

In real-time broadcast environments such as live sports coverage, break points are fluid depending on game situations, leading to issues with the labor burden of manual insertion methods or the timing inaccuracies of pre-scheduled methods. To address this, an automation architecture combining AWS Elemental MediaLive, Amazon Rekognition, and AWS Step Functions is introduced.

The core of this system is detecting Filler videos broadcast on the screen. Fillers are a definitive signal indicating the start of ad insertion intervals, such as halftime, and are fixed video assets, allowing AI to recognize them reliably.

The specific processing steps are as follows:

• Thumbnail Collection: Step Functions invokes Lambda at 1-second intervals to retrieve the current screen using MediaLive's DescribeThumbnails API.
• AI Analysis: Thumbnails are analyzed using Amazon Rekognition Custom Labels to determine if they are fillers.
• Signal Insertion: Upon detection, SCTE-35 signals are inserted into the stream via the BatchUpdateSchedule API.
• Evidence Storage: Detected thumbnails are saved to S3, and a cooldown is applied to prevent duplicate insertions.

This approach enables real-time screen sampling without a separate frame extraction pipeline and allows for stable maintenance of the ad monetization pipeline without manual operator intervention.