Available Objects | Frigate
Short summary
This doc explains how to configure which objects Frigate detects and tracks in video feeds, including how to filter false positives and mask detection zones. By default it only tracks people, but you can add cars, animals, and dozens of other object types.
- Frigate supports 90+ object types from Google Coral's dataset, but only tracks "person" by default — you must explicitly enable tracking for other objects like cars or dogs
- Object filters let you reduce false positives by setting minimum/maximum bounding box size, aspect ratio, and confidence thresholds per object type at global or per-camera level
- Mask zones prevent detection in specific frame areas (like trees or shadows) and can be applied globally or to individual object types
Full summary
Frigate detects 90+ object types pulled from Google Coral's training data, but only tracks people by default. The list includes obvious stuff like cars, dogs, and chairs, but also oddities like teddy bears and hair dryers. One quirk: "car" appears twice because the system renamed "truck" to "car" automatically, which the docs acknowledge causes confusion. If you want to track other objects—say, dogs or vehicles—you configure them either globally through the UI or per-camera in YAML. The configuration is straightforward: just add object names to a tracked list.
Filtering cuts down false positives by setting constraints on size, shape, and confidence scores for each object type. You can define minimum and maximum bounding box areas (in pixels or as a percentage of frame), width-to-height ratios, and confidence thresholds. Min score determines when Frigate starts tracking something; threshold is the confidence bar for considering it a real detection. These filters work globally or per-camera, so you can be strict about people in one camera feed and lenient in another. The system checks the bottom center of the bounding box against these rules.
Object filter masks let you block detections in specific frame areas—useful for ignoring motion near trees or parked cars that always trigger false alarms. Masks use normalized coordinates and apply either globally across all object types or just to specific ones. You can draw them directly in the UI on the live camera feed. The system also supports custom models if the bundled ones don't work for your setup; you mount your own TensorFlow Lite models for CPU or EdgeTPU processing and update the config accordingly.