Hybrid On-Device Cloud Scheme for Re-Identification of Persons Based on Shared Embedding Gallery

Generally, the present disclosure is directed to a system of facial and/or person recognition via machine learning and Internet of Things (IoT). In particular, in some implementations, the systems and methods of the present disclosure can include or otherwise leverage a machine learning and IoT system or device to track and/or identify a person based on video images taken by one or more device(s). For example, a hybrid on-device and cloud scheme can enable locally-derived embeddings from multiple camera devices to be sent to a shared cloud space which can cluster the embeddings to generate a person model for a given person. Later, a camera device participating in the scheme can again detect a face and can match an embedding generated for the face against the shared gallery of person models to (potentially) re-identify the previously observed person