Configuring Processors
Kraken Processors use different algorithms to process and analyze real-time network video. Internally, Kraken provides the decoded video, audio, and metadata essences to the configured processor, which then derives actionable insights, including AI object detection, from the source media.
Users can choose from an existing set of Kraken Processors to license within their Kraken instance. Notable processors include Sentient Tracker and the Yolo based image processor. Shield AI's Sentient Tracker can identify and track objects from airborne imagery in land and maritime environments, and the Yolo based image processer, packaged by Haivision, provides real-time object detection processing using the YOLOv8 algorithm. The number of compatible processors will continue to grow with time.
Currently, Processor Modules containing 3rd-party algorithms can only be integrated by Haivision. Contact your Haivision representative for more information about available Processor Modules.
Note
No processor modules are available on Kraken units by default, as processors are a licensed feature. For details on adding processor licenses and modules, see Adding New Processors.
Selecting a Processor
This section describes the Sentient Tracker Processor Module (algorithm by Shield AI) and the Yolo-based image processing Processor Module (YOLOv8 algorithm integrated by Haivision) to help you select the right type of processing for your stream.
Sentient Tracker Processor Module
Sentient Tracker leverages Shield AI's AI-enabled tracking to automatically detect moving objects within electro-optical (EO) or infrared (IR) full-motion video.
This processor is best suited for:
- ISR Missions in Land or Maritime environments,
- Search and Rescue (SAR) missions,
- Video processing from either crewed or uncrewed aircrafts,
- Detecting moving objects as small as 2x2 pixels.
The Sentient Tracker processor is compatible with compressed network streams, as well as raw SDI sources from gimbaled sensors common in airborne workflows.
Note
The Sentient Tracker processor is limited to only a single runtime per Kraken instance at this time.
Yolo Based Image Processing Module
This processor uses YOLOv8 advancements in deep learning and computer vision to provide fast and accurate video processing. This Processor Module serves as an example to demonstrate the possibilities in detection and annotation performance on different Kraken instance types.
This processor is best suited for:
- Object detection for land navigation,
- Security surveillance,
- Real-time distance tracking,
- Object counting (e.g. for inventory).
Yolo processors are suitable for various applications and hardware platforms, including edge devices and cloud APIs.
Stream Count Limitations
Significant resources are required to run streams with processors enabled, and overall stream throughput on a given Kraken instance may decrease when running one or more Kraken Processors.
Care is needed when planning Kraken deployments with significant real-time processing.
Generally, when a Kraken instance is licensed for real-time processing, the allowed runtime processor count will match the stream count for that instance. Some integrated real-time processors may have further simultaneous runtime restrictions imposed by the 3rd-party algorithm.