US 10,854,062 B2Granted

Fire monitoring system

AssigneeHochiki Corporation

InventorHiromichi Ebata

Life of this patent

Estimating the term…

Loading patent…

Abstract

A monitor image in a monitor region captured by a monitor camera is divided into a plurality of block images, which are inputted to a fire detector constituted by a multi-layer-type neural network so as to detect a fire. A learning image generation device divides a learning image in which a fire in the monitor region is captured into a plurality of block images and classifies the images to fire learning block images, normal learning block images, or non-learning block images based upon the presence/absence of flames and smoke and the ratio at which the flames and smoke occupy in each block image so as to be stored, and a learning control part inputs the fire learning block images and the normal learning block images stored in the learning image generation device to the fire detector so as to subject a multi-layer-type neural network to learning by deep learning process.

Claims8

Dependent claims

A fire monitoring system comprising:

an image capturing part for capturing a monitor image in a monitor region;
an image input part for dividing the monitor image into a plurality of block images as determining images by inputting the monitor image;
a multi-layer-type neural network for determining a flame block in which a flame is detected or a smoke block in which smoke is detected or a normal block in which neither a flame nor smoke is detected by inputting each of the block images as determining images;
a determination part for determining a fire in the monitor region based on the flame block and the smoke block;
a learning image generation part for dividing the monitor image serving into a plurality of block images as learning images and classifying each of the block images as learning images to a flame learning block image in which flames due to a fire occupy a ratio that is a predetermined ratio or more, a smoke learning block image in which smoke due to a fire occupies a ratio that is a predetermined ratio or more, a normal learning block image without flames or smoke, and a non-learning block image in which the proportion of flames and smoke is less than the predetermined proportion so as to be stored in a learning image storage part;
and
a learning control part for learning the multi-layer-type neural network by inputting the flame learning block image, smoke learning block image and the normal learning block image excluding the non-learning block image stored in the learning image storage part.

4 more elements

A fire monitoring system comprising:

an image capturing part for capturing a monitor image in a monitor region;
an image input part for dividing the monitor image into a plurality of block images as determining images by inputting the monitor image;
a first multi-layer-type neural network for determining a flame block in which a flame is detected or a normal block in which no flame is detected by inputting each of the block images as determining images;
a second multi-layer-type neutral network for determining a smoke block in which smoke is detected or a normal block in which no smoke is detected by inputting each of the block images as determining images;
a determination part for determining a fire in the monitor region based on the flame block and the smoke block;
a learning image generation part for dividing the monitor image serving into a plurality of block images as learning images and classifying each of the block images as learning images to a flame learning block image in which flames due to a fire occupy a ratio that is a predetermined ratio or more, a smoke learning block image in which smoke dues to a fire occupies a ratio that is a predetermined ratio or more, a normal learning block image without flames or smoke, and a non-learning block image in which the proportion of flames and smoke is less than the predetermined proportion so as to be stored in a learning image storage part;
and
a learning control part for learning the first multi-layer-type neural network by inputting the flame learning block image and the normal learning block image stored in the learning image storage part, and learning the second multi-layer-type neural network by for inputting the smoke learning block image and the normal learning block image stored in the learning image storage part.

5 more elements

Dependencies are inferred from the claim text.

Facts

Application: 16/411,665Filed 14 May 2019 · continuation
Priority: JP 2016-24740721 Dec 2016
Published: US 2019/0279478 A112 Sep 2019
Granted: 1 Dec 2020
Term ends: Not estimatedThe source records don’t support an estimate.

Classification16 codes

G08B 17/125
Actuation by presence of radiation or particles, e.g. of infrared radiation or of ions: by using a video camera to detect fire or smoke
G06N 3/045
Architecture, e.g. interconnection topology: combinations of networks
G06N 3/0464
Architecture, e.g. interconnection topology: convolutional networks [CNN, ConvNet]
G06N 3/084
Learning methods: backpropagation, e.g. using gradient descent