Core Definition
Re-Identification (Re-ID) is a computer-vision technique that determines whether objects seen by different cameras are the same object, using appearance-feature embeddings. It does not reveal names or identity — it anonymously links "A on camera 3" with "B on camera 7," and is a core building block of multi-camera tracking (MTMC).
The misunderstanding behind the name "Re-Identification"
Going by the term alone, it sounds like technology that figures out who someone is, but Re-ID in computer vision is different from identification. Where face recognition aims to answer "this person is Jane Doe," Re-ID aims to answer "the object from a moment ago is this object now." It joins fragments together without ever attaching a name.
Running Re-ID directly on raw footage, however, leaves room for it to be combined with identity information. That is why the choice of what data it runs on (raw footage or anonymized features) becomes the fork in the road for privacy.
What anonymous Re-ID makes possible
Anonymized features alone are enough for store analysis. Analyses such as the path from entry to exit, movement between zones, and repeat-visit patterns only need to know "the same object," not "who" it is.
When this distinction is maintained, precise flow analysis across multi-store operations and large spaces can coexist with privacy protection.
How SAAI Uses It
DEEPINGSOURCE's Re-ID uses only anonymized features. Identity elements such as faces are already removed at the point of input, and only the remaining appearance-feature embeddings are used to link movement. The embeddings are designed so that recovering a person's identity from them is impossible: the system can tell that it is "the same object," but "who" that object is exists nowhere in the system.
See it in action