Flame finding with automated image analysis

Inventors

Bonn, David

Assignees

Deep Seek Labs Inc

Publication Number

US-11145090-B2

Publication Date

2021-10-12

Expiration Date

Interested in licensing this patent?

MTEC can help explore whether this patent might be available for licensing for your application.


Abstract

A computer system determines a likelihood of the presence of flame in a scene. The system obtains a series of digital infrared and optical images of the scene; identifies a candidate region in a location of the scene based on analysis of the infrared images;identifies an optical image slice based on analysis of the optical images and the location of the candidate region; and determines a likelihood of the presence of flame in the scene based on analysis of the optical image slice. Analysis of the infrared images includes detecting a high-temperature image region that exceeds a threshold temperature; detecting a turbulent motion image region; and determining whether the turbulent motion region is within a specified proximity of or overlaps with the high-temperature region. The optical image slice may be provided to a trained neural network, which returns a degree-of-confidence value that indicates whether flame is present.

Core Innovation

A computer system determines a likelihood of the presence of flame in a scene by obtaining a series of digital infrared images and digital optical images of a scene; identifying a candidate region in a location of the scene based at least in part on analysis of the series of digital infrared images; identifying an optical image slice within an optical image region corresponding to the location of the candidate region based at least in part on analysis of the series of digital optical images; and determining a likelihood of the presence of flame in the scene based at least in part on analysis of the optical image slice. Analysis of the series of digital infrared images includes detecting one or more high-temperature image regions in the series of digital infrared images that exceed a threshold temperature; detecting one or more turbulent motion image regions in the series of digital infrared images; and determining whether a turbulent motion image region is within a specified proximity of or overlaps with a high-temperature image region.

The disclosure addresses shortcomings of optical-only flame detection by combining infrared analysis (hot spots and turbulent motion) with optical image evaluation and machine learning. Optical image slices corresponding to infrared candidate regions can be passed as input to a trained neural network that returns a degree-of-confidence value indicating whether flame is present, and the system can perform further actions such as setting an alarm state or transmitting signals based on the determined likelihood.

Claims Coverage

This patent includes three independent claims presenting inventive features related to detecting flame using combined infrared and optical image analysis and a system implementation.

Likelihood determination using infrared and optical images

Obtaining a series of digital infrared images and digital optical images of a scene; identifying a candidate region in a location of the scene based at least in part on analysis of temperature and turbulent motion in the series of digital infrared images; identifying an optical image slice within an optical image region corresponding to the location of the candidate region based at least in part on analysis of the series of digital optical images; and determining a likelihood of the presence of flame in the scene based at least in part on analysis of the optical image slice.

Identifying candidate regions in infrared images based on temperature and motion

Obtaining a series of digital infrared images of a scene; detecting one or more high-temperature image regions in the scene that exceed a threshold temperature based on analysis of the series of digital infrared images; detecting one or more turbulent motion image regions in the scene based on analysis of the series of digital infrared images; and identifying one or more candidate regions for location of flame in the series of digital infrared images based at least in part on locations of the one or more high-temperature image regions and the one or more turbulent motion image regions.

System implementing combined infrared and optical analysis

A system comprising one or more computing devices programmed to determine a likelihood of the presence of flame in a scene by performing operations comprising: obtaining a series of digital infrared images and digital optical images of a scene; identifying a candidate region in a location of the scene based at least in part on analysis of temperature and turbulent motion in the series of digital infrared images; identifying an optical image slice within an optical image region corresponding to the location of the candidate region based at least in part on analysis of the series of digital optical images; and determining a likelihood of the presence of flame in the scene based at least in part on analysis of the optical image slice.

Independent claims cover (1) a method for determining flame likelihood by combining infrared temperature and turbulent motion analysis with optical image slice analysis, (2) infrared-only candidate region identification based on high-temperature and turbulent motion detection, and (3) a system programmed to perform the combined infrared and optical analysis.

Stated Advantages

Provides early warning of fires, such as wildfires, by image-based flame detection.

Improves processing times by restricting areas of the optical image that need to be analyzed using infrared candidate regions.

Reduces the risk of false positives by using infrared detection of hot spots and turbulent motion to distinguish fire from other bright features.

Modified SqueezeNet provides good accuracy with a compact memory footprint of approximately 6 MB.

System can remain functional in local mode when an Internet connection is unavailable to activate local flame detection alarms.

Documented Applications

Early warning of fires, such as wildfires, for fire departments, government agencies, and residents of the wildland-urban interface.

Setting an alarm state and initiating actions based on determined likelihood of flame, including activating a visual or audible alarm and activating a fire suppression system.

Transmitting signals to a remote computing device and generating automated messages or phone calls based on detected flame likelihood.

Causing information to be displayed on a roadway variable-message sign in response to detected flame.

JOIN OUR MAILING LIST

Stay Connected with MTEC

Keep up with active and upcoming solicitations, MTEC news and other valuable information.