What happens in the backend when two identities are merged?

Written by NSoft Vision


Each detected identity is stored in a database on a central platform as a vector. Each vector is unique and is a direct numerical indicator of biometric data collected from the face of the subject. The value of the vector is not randomized in any sense. 

There is a new solution in the Vision system that happens in the backend once identities are manually merged. The new way merging is implemented so that the vectors of identities that are being merged are always added to collection of new identity vectors, regardless of the distance between vectors of identities that are in merge process.

Therefore, newly created identity gets the collection of vectors and each vector from that collection is used for future recognition and is not mutually interfering with one another.

The important thing to understand is that the system does not merge any types of images or sightings, but vectors. The way it perceives identities is through the value of the vector.

Before doing any sort of manual merge it's important to consider the distance view shown on GUI under "Similar Identities". The closer the distance view of a similar identity is to the threshold of 0.6 more similar the vectors are of those two compared identities. Further away from that threshold the value is, bigger the distance of their vectors is.