Skip to main content
Version: V12

Understanding the Object Library

Detection tells you that something is in frame, a person, a vehicle, a face. The Object Library is what tells you who or which one. It is your portal's gallery of known subjects, the people, vehicles, and objects you have enrolled ahead of time, so that when a camera sees one of them, the system doesn't just report "a person," it reports the specific subject you are watching for.

This article explains what the Object Library is, how a subject is built from reference photos, and how enrolled subjects turn into recognitions you can act on.

What the Object Library Is​

The Object Library is a per-portal recognition gallery, sometimes thought of as a watchlist. You enroll a subject by giving it a name and uploading one or more reference photos. From each photo, the AI service builds a mathematical signature of what the subject looks like. While your cameras are live, every detection is compared against that gallery, and when a detection is close enough to an enrolled subject, it becomes a recognition.

So the Object Library sits one layer above detection. Detection spots the object; recognition matches it to a subject you already care about.

Subjects and Reference Photos​

A subject is one person, vehicle, or object you want recognized. Each subject holds:

  • A Name, and optionally an External ID that ties it to a record in another system (a case number, an employee ID, a license record).
  • A Type, the kind of subject it is, chosen when you enroll it.
  • A Watchlist Priority, how much it matters relative to other subjects.
  • A Description and Tags for organizing and searching the gallery.
  • One or more reference photos, the images the system learns the subject's appearance from.

A subject can hold several photos, and more good photos generally means more reliable recognition, different angles, lighting, and distances give the system more to match against. All photos under a subject describe the same individual, so a recognition of any of them counts as recognizing that one subject.

Subject Types​

Every subject has a Type, and the type decides which recognition pipeline processes its photos:

  • Face: recognize a specific person by their face.
  • Person: recognize a specific individual by their overall appearance.
  • Vehicle: recognize a specific vehicle.
  • Object: recognize a specific object.

Important: A subject's type is fixed once it is enrolled, because the photo's signature is produced by that type's pipeline. You can edit a subject's name, description, priority, and tags later, but not its type. If you picked the wrong type, delete the subject and enroll it again.

Note: Live recognition currently runs for faces. You can organize subjects under any of the four types, but the type matched against live camera feeds today is Face. Recognition for Person, Vehicle, and Object subjects is planned, and the library is built on a common framework so those models can plug into the same enrollment and matching flow as they become available. See Understanding Face Recognition.

How a Photo Becomes a Reference​

When you add a photo, the AI service doesn't just store the image. It detects the relevant object in the photo, draws a bounding box around the one it will learn from, and scores the result on two dimensions:

  • Detection score, how confidently it found the subject in the photo.
  • Quality score, how usable that detected region is as a reference, factoring in things like size, sharpness, and angle.

A clear, well-lit, head-on photo earns a high quality score and makes a strong reference. A small, blurry, or side-on subject earns a low one. Low-quality references can drag recognition accuracy down, so by default the system steers you toward good photos. When you genuinely need to enroll from a weaker image, you can choose to accept marginal quality and keep it anyway.

Watchlist Priority​

Watchlist Priority, High, Medium, or Low, marks how much each subject matters. It does not change whether a subject is recognized; it ranks what surfaces first when a recognition happens, so a high-priority subject stands out against routine matches. Set it to reflect real-world stakes: a person of interest sits above a routine visitor.

From Enrollment to Recognition​

Once a subject is enrolled, the recognition loop runs on its own:

  1. A camera detects an object, a face, a person, a vehicle.
  2. The detection is compared against the subjects of that type in the Object Library.
  3. If it matches an enrolled subject closely enough, it is recorded as a recognition of that subject.

Each recognition is captured as activity on the subject. In the subject's detail view you can see when and where it was last seen, browse the recordings it appeared in, and jump straight to the moment in the footage. A subject that has not been matched yet simply shows no activity.

Use Cases​

Person of Interest Watchlist​

A security team enrolls known individuals as Face subjects with High priority. When any camera sees a match, the recognition is logged and the operator is pointed to the recording, turning hours of passive footage into a short list of moments that matter.

Fleet and Vehicle Tracking (planned)​

When Vehicle recognition ships, a facility will be able to enroll its own vehicles, or vehicles of interest, as Vehicle subjects, so recognitions across entrances and lots build a record of where and when each vehicle appeared, without anyone scrubbing through video. Today this scenario applies to Face subjects; vehicle matching is planned.

Key Considerations​

  • Photo quality drives accuracy. A few strong reference photos beat many weak ones. Prefer clear, well-lit images where the subject is the obvious focus.
  • Type cannot be changed after enrollment. Choose it deliberately; correcting it means re-enrolling.
  • Recognition depends on detection. A subject is only matched when its type is actually being detected on the camera that sees it. Pair the library with the right detection settings per camera.
  • The gallery is per portal. Subjects are isolated to your portal and are not shared across tenants.
  • Access is permission-gated. Viewing, adding, editing, and deleting subjects are governed by the Object Library permissions on your group or CAL.

See Also​

Did you find this article helpful?