Verfahren und Vorrichtung zur Erkennung einer Anomalie in einem Video

Details

Veröffentlichungs-Nr.

EP4383213

Aktenzeichen

EP22306803

Anmeldetag

7. Dezember 2022

Veröffentlichung

11. März 2026

Erteilung

11. März 2026

Rechtsraum

EP

IPC

G06V20/40G06V20/52

Abstract

The present invention relates to a method and device for detecting an anomaly in a video (VID) using a neural network (HSN) comprising a first neural sub-network (HN), a second neural sub-network (SN), and a coupler (CPL). In particular, the proposed method comprises: - obtaining (S100), using the first neural sub-network (HN), first anomaly scores (HS<1>-HS<T>) for a plurality of temporal segments (t<1>-t<T>) of the video (VID), first anomaly scores (HS<1>-HS<T>) being representative of anomalous human activities; - obtaining (S200), using the second neural sub-network (SN), second anomaly scores (SS<1>-SS<T>) for the temporal segments (t<1>-t<T>) representative of anomalous events in the scene; and - obtaining (S300), using the coupler (CPL), aggregated anomaly scores (AS<1>-AS<T>) for the temporal segments (t<1>-t<T>) based on a weighted sum of the first and second anomaly scores (HS<1>-HS<T>, SS<1>-SS<T>).

Anmelder

Firmen

TOYOTA JIDOSHA KABUSHIKI KAISHA
Institut National de Recherche en Informatique et en Automatique

Land

🇯🇵 Japan

Fachgebiete