Self-learning video fire detection method

Abstract

The invention discloses a self-learning video fire detection method. According to the technical field, the method comprises the steps as follows: a batch of fireworks images are collected, the backgrounds of the images are removed and tags are added, and the images are taken as a seed data set; data samples for a specific application environment are automatically generated using the seed data and a collected video; the data samples are trained through an online learning algorithm to get fire and smoke detectors; a front end sends an acquired suspected fire image to a detection server to judge whether there is a fire; and in the running process, the detection server gets new samples constantly to update the detectors online. Compared with the prior art, the method has the following advantages: the method can be used in all kinds of scenarios, and can adaptively improve the performance and reduce the rate of false report; and suspected area extraction and fire identification are separated, and fire identification can be conducted remotely or even be deployed at the cloud, which makes the whole system more convenient to upgrade and maintain.

Description

technical field [0001] The invention relates to the fields of fire detection and video monitoring, in particular to a self-learning video fire detection method in the fields of fire detection and video monitoring. Background technique [0002] Image-based fire detectors based on video image processing technology can detect fires in the early stages of fires, and can intuitively respond to on-site conditions, which is conducive to early detection and early extinguishment of fires, especially in applications such as outdoor and indoor tall spaces. , has received more and more attention, and the application demand is strong. [0003] The core of the image-based fire detector is the image analysis algorithm, which requires the ability to accurately identify flames and smoke, without missing or false alarms. The image-based fire detection system generally adopts a structure in which the data collected by the front-end camera is transmitted to the back-end server for fire identif...

AI Technical Summary

Problems solved by technology

It is extremely difficult to obtain an image fire detector that can be assembled in various occasions. Even if the identification module parameters are configured for specific occasions during deployment, during operation, due to factors such as seasons, weather, and piles, the equipment has to deal with The environment has also changed, which will gradually lead to prominent false positives and difficult system maintenance.

Images