FLAME IDENTIFICATION METHOD AND DEVICE USING IMAGE ANALYSES IN HSI COLOR SPACE

• US 20120195462A1
• Filed: 09/09/2011
• Published: 08/02/2012
• Est. Priority Date: 01/27/2011
• Status: Active Grant

• Current Assignee: Chang Jung Christian University
• Original Assignee: Chang Jung Christian University
• Inventors: PU, Yong-Ren, LEE, Su-Hsing, CHEN, Yung-Jen
• Granted Patent: US 8,538,080 B2
• Time in Patent Office: Days
• US Class Current: 382/103
• CPC Class Codes: G06T 2207/10024 Color image
• CPC Class Codes: G06T 2207/30232 Surveillance
• CPC Class Codes: G06T 7/136 involving thresholding
• CPC Class Codes: G06T 7/194 involving foreground-background segmentation
• CPC Class Codes: G06T 7/90 Determination of colour characteristics
• CPC Class Codes: G08B 17/125 by using a video camera to detect fire or smoke
• CPC Class Codes: G08B 29/185 Signal analysis techniques for reducing or preventing false alarms or for enhancing the reliability of the system

Abstract

In a flame identification method and device for identifying any flame image in a plurality of frames captured consecutively from a monitored area, for each image frame, intensity foreground pixels are obtained based on intensity values of pixels, a fire-like image region containing the intensity foreground pixels is defined when an intensity foreground area corresponding to the intensity foreground pixels is greater than a predetermined intensity foreground area threshold, and saturation foreground pixels are obtained from all pixels in the fire-like image region based on saturation values thereof to obtain a saturation foreground area corresponding to the saturation foreground pixels. Linear regression analyses are performed on two-dimensional coordinates each formed by the intensity and saturation pixel areas associated with a corresponding image frame to generate a determination coefficient. Whether a flame image exists in the image frames is determined based on the determination coefficient and a predetermined identification threshold.

Claims

1. A flame identification method of identifying any flame image in a plurality of image frames captured consecutively from a monitored area, each of image frames including a plurality of pixels each having pixel data converted into a hue value, a saturation value and an intensity value in hue-saturation-intensity (HSI) color space, said flame identification method comprising the steps of:

a) for each of the image frames,

filtering the pixels based on the intensity values corresponding to the pixels, and intensity reference information to divide the pixels into intensity foreground pixels associated with fire-like intensity performance and intensity background pixels so as to obtain an intensity foreground area corresponding to the intensity foreground pixels, and so as to define within a corresponding one of the image frames a fire-like image region that contains the intensity foreground pixels when the intensity foreground area is greater than a predetermined intensity foreground area threshold,

filtering all pixels in the fire-like image region based on the saturation values corresponding to all the pixels in the fire-like image region, and saturation reference information to divide all the pixels in the fire-like image region into first saturation foreground pixels associated with weak-fire saturation performance and saturation background pixels so as to obtain a first saturation foreground area corresponding to the first saturation foreground pixels, and

combining the intensity foreground area and the first saturation foreground area associated with the corresponding one of image frames to form a first two-dimensional (2-D) coordinate;

b) when a number (N) of the first 2-D coordinates corresponding to a sequence of the image frames are formed consecutively in step a), performing linear regression analyses on the first 2-D coordinates to generate a first determination coefficient; and

c) determining whether a flame image exists in the image frames based on the first determination coefficients generated in step b) and a predetermined identification threshold.

2. The flame identification method as claimed in claim 1, wherein, in step c), when the first determination coefficient is greater than the predetermined identification threshold, it is determined that a flame image exists in the image frames.

3. The flame identification method as claimed in claim 1, prior to step b), further comprising the step of:

b-0) for the fire-like image region of the corresponding one of the image frames, obtaining from the first saturation foreground pixels obtained in step a) second saturation foreground pixels, which do not overlap the intensity foreground pixels of the corresponding one of the image frames obtained in step a), so as to obtain a second saturation foreground area corresponding to the second saturation foreground pixels, and combining the intensity foreground area and the second saturation foreground area associated with the corresponding one of the image frames to form a second 2-D coordinate;

wherein, in step b), when a number (N) of the second 2-D coordinates are formed consecutively in step b-0), linear regression analyses are further performed on the second 2-D coordinates to generate a second determination coefficient; and

wherein, in step c), whether a flame image exists in the image frames is determined further based on the second determination coefficient generated in step b).

4. The flame identification method as claimed in claim 3, wherein, in step c), it is determined that a flame image exists in the image frames when each of the first and second determination coefficients is greater than the predetermined identification threshold.

5. The flame identification method as claimed in claim 1, wherein in step a):

the intensity reference information includes a predetermined intensity foreground threshold, a predetermined intensity variation lower limit value (σ

_(I)min), a predetermined intensity amplification ratio (B_(I)), a predetermined intensity weight (β

_(I)), and a plurality of sets of intensity average values ( P_(I)), intensity average variation values ( σ

_(I)) and intensity average variation thresholds ( σ

_{(I)threshold)}, each set of the intensity average value ( P_(I)), the intensity average variation value ( σ

_(I)) and the intensity average variation threshold ( σ

_{(I)threshold)} being preset previously for a corresponding pixel;

for each of the intensity foreground pixels divided from the pixels of the corresponding one of the image frames,

the intensity value is greater than the predetermined intensity foreground threshold, and

a difference between the intensity value, and a corresponding intensity average value ( P_(I)) is greater than a product of a corresponding intensity average variation threshold ( σ

_{(I)threshold)} and the predetermined intensity amplification ratio (B_(I)); and

for each of the intensity background pixels divided from the pixels of the corresponding one of the image frames,

a corresponding intensity average variation value ( σ

_(I)) is updated with a sum of a product of the predetermined intensity weight (σ

_(I)) and a difference between the intensity value and the corresponding intensity average value ( P_(I)), and a product of the corresponding intensity average variation value ( σ

_(I)) and a difference between 1 and the predetermined intensity weight (β

_(I)),

the corresponding intensity average value ( P_(I)) is updated with a sum of a product of the predetermined intensity weight (β

_(I)) and the intensity value, and a product of the corresponding intensity average value ( P_(I)) and the difference between 1 and the predetermined intensity weight (β

_(I)), and

a corresponding intensity average variation threshold ( σ

_{(I)threshold)} is up dated with a larger one of the predetermined intensity variation lower limit value (σ

_(I)min) and the corresponding intensity average variation value ( σ

_(I)).

6. The flame identification method as claimed in claim 1, wherein in step a):

the saturation reference information includes a predetermined saturation lower limit value, a predetermined saturation upper limit value, a predetermined saturation variation lower limit value (σ

_(S)min), a predetermined saturation amplification ratio (B_(S)), a predetermined saturation weight (β

_(S)), and a plurality of sets of saturation average values ( P_(S)) saturation average variation values ( σ

_(S)) and saturation average variation thresholds ( σ

_(S)threshold), each set of the saturation average value ( P_(S)), the saturation average variation value ( σ

_(S)) and the saturation average variation threshold ( σ

_(S)threshold) being preset previously for a corresponding pixel;

for each of the first saturation foreground pixels divided from all the pixels in the fire-like image region of the corresponding one of the image frames,

the saturation value is within a range from the predetermined saturation lower limit value to the predetermined saturation upper limit value, and

a difference between the saturation value, and a corresponding saturation average value ( P_(S)) is greater than a product of a corresponding saturation average variation threshold ( σ

_(S)threshold) and the predetermined saturation amplification ratio (B_(S)); and

for each of the saturation background pixels divided from all the pixels in the fire-like image region of the corresponding one of the image frames,

a corresponding saturation average variation value ( σ

_(S)) is updated with a sum of a product of the predetermined saturation weight (β

_(S)) and a difference between the saturation value ( P_(S)) and the corresponding saturation average value ( P_(S)), and a product of the corresponding saturation average variation value ( σ

_(S)) and a difference between 1 and the predetermined saturation weight (β

_(S)),

the corresponding saturation average value ( P_(S)) is updated with a sum of a product of the predetermined saturation weight (β

_(S)) and the saturation value, and a product of the corresponding saturation average value ( P_(S)) and the difference between 1 and the predetermined saturation weight (β

_(S)), and

a corresponding saturation average variation threshold ( σ

_(S)threshold) updated with a larger one of the predetermined saturation variation lower limit value (σ

_(S)min) and the corresponding saturation average variation value ( σ

_(S)).

7. A flame identification device for identifying any flame image in a plurality of image frames captured consecutively from a monitored area, each of the image frames including a plurality of pixels, each of which has pixel data, said flame identification device comprising:

a pixel data processor for converting the pixel data of each of the pixels of each of the image frames into a hue value, a saturation value and an intensity value in the hue-saturation-intensity (HSI) color space;

a memory unit for storing intensity reference information and saturation reference information; and

a processing unit coupled to said pixel data processor and said memory unit, and receiving the intensity values and the saturation values corresponding to each of the image frames from said pixel data processor, and including

an intensity processor configured to filter, for each of the image frames, the pixels based on the intensity values corresponding to the pixels, and the intensity reference information from said memory unit to divide the pixels into intensity foreground pixels associated with fire-like intensity performance and intensity background pixels so as to obtain an intensity foreground area corresponding to the intensity foreground pixels, to determine whether the intensity foreground area is greater than a predetermined intensity foreground area threshold, and to define within a corresponding one of the image frames a fire-like image region that contains the intensity foreground pixels upon detecting that the intensity foreground area is greater than the predetermined intensity foreground area threshold, and

a saturation processor configured to filter all pixels in the fire-like image region defined by said intensity processor based on the saturation values corresponding to all the pixels in the fire-like image region, and the saturation reference information from said memory unit to divide all the pixels in the fire-like image region into first saturation foreground pixels associated with weak-fire saturation performance and saturation background pixels so as to obtain a first saturation foreground area corresponding to the first saturation foreground pixels, and

a determination processor configured to combine the intensity foreground area and the first saturation foreground area associated with the corresponding one of the image frames to form a first two-dimensional (2-D) coordinate, to perform linear regression analyses on a number (N) of the first 2-D coordinates formed consecutively thereby to generate a first determination coefficient, and to determine whether a flame image exists in the image frames based on the first determination coefficient and a predetermined identification threshold.

8. The flame identification device as claimed in claim 7, wherein, when the first determination coefficient is greater than the predetermined identification threshold, said determination processor determines that a flame image exists in the image frames.

9. The flame identification device as claimed in claim 7, wherein:

for the fire-like image region of the corresponding one of the image frames, said saturation processor further obtains, from the first saturation foreground pixels, second saturation foreground pixels, which do not overlap the intensity foreground pixels of the corresponding one of the image frames obtained by said intensity processor, so as to obtain a second saturation foreground area corresponding to the second saturation foreground pixels; and

said determination processor further combines the intensity foreground area and the second saturation foreground area associated with the corresponding one of the image frames to form a second 2-D coordinate, performs linear regression analyses further on a number (N) of the second 2-D coordinates formed consecutively thereby to generate a second determination coefficient, and determines whether a flame image exists in the image frames further based on the second determination coefficient.

10. The flame identification device as claimed in claim 9, wherein said determination processor determines that a flame image exists in the image frames when each of the first and second determination coefficients is greater than the predetermined identification threshold.

11. The flame identification device as claimed in claim 7, wherein:

the intensity reference information stored in said memory unit includes a predetermined intensity foreground threshold, a predetermined intensity variation lower limit value (σ

_(I)min), a predetermined intensity amplification ratio (B_(I)), a predetermined intensity weight (β

_(I)), and a plurality of sets of intensity average values ( P_(I)), intensity average variation values ( σ

_(I)) and intensity average variation thresholds ( σ

_(I)threshold), each set of the intensity average value ( P_(I)), the intensity average variation value ( σ

_(I)) and the intensity average variation thresholds ( σ

_(I)threshold) being preset previously for a corresponding pixel;

for each of the intensity foreground pixels divided by said intensity processor from the pixels of the corresponding one of the image frames,

the intensity value is greater than the predetermined intensity foreground threshold from said memory unit, and

a difference between the intensity value, and a corresponding intensity average value ( P_(I)) from said memory unit is greater than a product of a corresponding intensity average variation threshold ( σ

_(I)threshold) and the predetermined intensity amplification ratio (B_(I)) from said memory unit; and

for each of the intensity background pixels divided by said intensity processor from the corresponding one of the image frames,

said intensity processor updates a corresponding intensity average variation value ( σ

_(I)) stored in said memory unit with a sum of a product of the predetermined intensity weight (β

_(I)) from said memory unit and a difference between the intensity value, and the corresponding intensity average value ( P_(I)) from said memory unit, and a product of the corresponding intensity average variation value ( σ

_(I)) from said memory unit and a difference between 1 and the predetermined intensity weight (β

_(I)) from said memory unit,

said intensity processor updates the corresponding intensity average value ( P_(I)) stored in said memory unit with a sum of a product of the predetermined intensity weight (β

_(I)) from said memory unit and the intensity value, and a product of the corresponding intensity average value ( P_(I)) from said memory unit and the difference between 1 and the predetermined intensity weight (β

_(I)) from said memory unit, and

said intensity processor updates a corresponding intensity average variation threshold ( σ

_(I)threshold) stored in said memory unit with a larger one of the predetermined intensity variation lower limit value (σ

_(I)min) and the corresponding intensity average variation value ( σ

_(I)) from said memory unit.

12. The flame identification device as claimed in claim 7, wherein:

the saturation reference information stored in said memory unit includes a predetermined saturation lower limit value, a predetermined saturation upper limit value, a predetermined saturation variation lower limit value (σ

_(S)min), a predetermined saturation amplification ratio (B_(S)), a predetermined saturation weight (β

_(S)) and a plurality of sets of saturation average values ( P_(S)), saturation average variation values ( σ

_(S)) and saturation average variation thresholds ( σ

_(S)threshold), each set of the saturation average value ( P_(S)), the saturation average variation value ( σ

_(S)) and the saturation average variation threshold ( σ

_(S)threshold) being preset previously for a corresponding pixel;

for each of the first saturation foreground pixels divided by said saturation processor from the fire-like image region of the corresponding one of the image frames,

the saturation value is within a range from the predetermined saturation lower limit value from said memory unit to the predetermined saturation upper limit value from said memory unit, and

a difference between the saturation value, and a corresponding saturation average value ( P_(S)) from said memory unit is greater than a product of a corresponding saturation average variation threshold ( σ

_(S)threshold) and the predetermined saturation amplification ratio (B_(S)) from said memory unit; and

for each of the saturation background pixels divided by said saturation processor from the fire-like image region of the corresponding one of the image frames,

said saturation processor updates a corresponding saturation average variation value ( σ

_(S)) stored in said memory unit with a sum of a product of the predetermined saturation weight (β

_(S)) from said memory unit and a difference between the saturation value, and the corresponding saturation average value ( P_(S)) from said memory unit, and a product of the corresponding saturation average variation value ( σ

_(S)) from said memory unit and a difference between 1 and the predetermined saturation weight (β

_(S)) from said memory unit,

said saturation processor updates the corresponding saturation average value ( P_(S)) stored in said memory unit with a sum of a product of the predetermined saturation weight (β

_(S)) from said memory unit and the saturation value, and a product of the corresponding saturation average value ( P_(S)) from said memory unit and the difference between 1 and the predetermined saturation weight (β

_(S)) from said memory unit, and

said saturation processor updates a corresponding saturation average variation threshold ( σ

_(S)threshold) stored in said memory unit with a larger one of the predetermined saturation variation lower limit value (σ

_(S)min) from said memory unit and the corresponding saturation average variation value ( σ

_(S))from said memory unit.