Kangal.

A camera can record an approaching predator. Someone still has to notice it. Kangal explores the connection between animal detection and a clear, reviewable alert for a farm operator.

The demonstration follows a fox in meadow footage, with a separate poultry-enclosure recording to show the protected subject. The recordings come from different locations; the farm scenario and watch zones are illustrative.

Google OWL-ViT analyses video frames and returns animal locations and model scores. Kangal links adjacent detections into tracks and evaluates the operator’s score, duration and watch-zone rules.

A sustained fox detection creates an incident, draws a red box around the animal and sounds an enabled alarm. The operator can acknowledge it, inspect a captured frame, replay the observation and resolve the review.

The video and detection results share one timeline. Replay, seeking and resizing preserve their alignment. Incident replay remains silent, while mute, acknowledgement and resolution have distinct effects.

Original sources, footage licences, model revision and downloadable analysis are available inside the demonstration. Inference is performed before playback; no live camera connection or production farm deployment is claimed.

The intended benefit is less time spent watching uneventful footage and a faster route to the moment that needs attention. Those benefits are proposed, rather than measured.

A farm deployment would need suitable fixed cameras, representative site footage and validation across lighting, weather, distance and species before unattended operation.