The invention relates to a
flame detection method based on multispectral videos. The core technology of the
flame detection method is that each
spectrum band of the visual spectrum is registered automatically, by means of the spectrum characteristic and the video motion characteristic, three
flame essential attributes including heat emitting, characteristic peaks and
flicker are detected, and the detection results of the three flame essential attributes are effectively fused. According to the invention, multispectral micro-
lens array video spectrum data serve as input, and the
flame detection method comprises an
algorithm used for registering the spatial position of each
spectrum band of the video spectrum image, a multispectral characteristic peak identification
algorithm used for distinguishing flame and another object having a color being similar to that of flame, a
flicker characteristic detection
algorithm used for distinguishing flame and luminous bodies such as incandescent lamps and the like; and an
information fusion algorithm used for fusing the multispectral characteristic with the
flicker characteristic. The
flame detection method based on multispectral videos can be used for detecting the early-stage
fire behavior of areas such as warehouses,
machine rooms and workshops, and can cooperate with an automatic alarm to notify the fire fighting department timely, and the
fire behavior can be prevented from being enlarged effectively.