A firework identification alarm method and system based on a multi-modal large model

Abstract

The application discloses a firework identification and alarm method and system based on a multi-modal large model: obtaining a to-be-identified video stream and an end-side distillation model; inputting an image frame in the to-be-identified video stream into the end-side distillation model to obtain a firework preliminary screening result corresponding to the image frame; when the firework preliminary screening result meets a preset firework condition, obtaining continuous multiple image frames containing a candidate firework target, and determining a time sequence dynamic feature of the candidate firework target based on a target position change of the candidate firework target in the continuous multiple image frames; judging whether the candidate firework target meets a preset firework dynamic consistency condition according to the time sequence dynamic feature, and if yes, sending the continuous multiple image frames to the cloud for review; and outputting an alarm instruction according to a review result of the cloud. The application realizes firework identification and alarm with high recall, low false alarm, low delay and low bandwidth in an end-side limited scene.

Images