Research Project Presentation

FIRE DETECTION USING GRAYSCALE VIDEO PROCESSING FOR BLACK AND WHITE

CAMERASby

GAURAV KISHOR DESHMUKH

Leonard Brown, PhD., Committee ChairDepartment of Computer Science

May 1, 2023Research Project Presentation

Research Committee:Dr. Leonard Brown IIIDr. Arun KulkarniDr. Naryanan Subramanian

Overview

Motivation Literature Review Proposed System Overview Performance Evaluation Conclusion


Motivation

May 1, 2023

Human and Economic Loss

Reasons for failure of Traditional Fire sensors Locality of fire Fail to detect potential range of fire

Visually Detectable Fire Characteristics Fire Color Motion Self Luminosity Flickering Smoke

Research Project Presentation

Overview




Literature Review Core Component of the color based fire detection techniques

Fire has yellowish - red color

Various color models to analyze colors in video sequences or images Red Green Blue (RGB)

• Concentrates more on red color• Ignores brightness • Potentially can detect any red object as fire• Is based on experimental threshold

May 1, 2023Graduate Seminar Presentation

Literature Review Hue Saturation Intensity (HSI)

• Concentrates more on hues• Considers brightness but ignores luminosity• Is based on experimental threshold

YCbCr • Considers luminosity• Luminosity is a factor of basic Green (G)• Can potentially detect any greenish object as fire


Literature Review Mixture Models

• Are composed of combinations of color model or mathematical transforms of same

• Gaussian RGB mixtureo Constructed using frequency transform on RGBo Improves hue of fire o Fails to consider self luminosity feature

• RGB + HSIo Others hues of red and brightness are consideredo Is based on threshold

Literature Review Flickering is another important characteristic of fire It distinguishes the real fire from image of fire Motion can be detected in video sequences using :

Temporal Differencing• Consecutive frames in the videos are subtracted to detect

motion• Detects every different pixel from the video


Literature Review Background Differencing

• Static background is subtracted from number of frames• Detects only moving pixel , fails to give sense of direction

Optical Flow Method• Motion and direction of each and every pixel in the frame is

considered• With the direction , Potential range of fire spread can be

calculated


Literature Review In any detection algorithm , False rate needs to be minimized For minimization some kind of decision mechanism is used Decision mechanisms surveyed are :

Neural Networks• Trained neural networks reduces rate of false detected pixels

drastically Fuzzy Logic

Rule based detection will help us determine true fire pixel Decision mechanisms are supportive

• Does not work if input given to it is faultyMay 1, 2023Graduate Seminar Presentation

Overview



Proposed System Overview


CCTV VideoB/W

Conversion Module

Fire Detector Module

Alarm



CCTV Video

B/W Conversion

Module


Alarm



Video I/P

CCTV Video

Module

Multiple FramesSingle Frame

To next module

CCTV Video Module



CCTV Video Module



CCTV VideoB/W

Conversion Module


Alarm



cvCvtColor() function

Frame

Pixel in given frameRGB Components

Corresponding gray pixel

B/W Conversion Module



B/W Conversion Module

B/W Conversion

Module



CCTV VideoB/W

Conversion Module


Alarm



Fire Detector Module Frame Analysis Pixel Analysis Classifier



Fire Detector Module Frame Analysis

Background Subtraction

Frame (N) Frame Difference

= - Frame (N-1)



Background Subtraction

fire pixels in motion




Background Subtraction Contour Analysis



Contour Analysis

Boundary of pixels in motion




• Output of overall frame

analysis is obtained by performing

AND operation between frames

obtained from background subtraction

and contour analysis component



Fire Detector Module Pixel Analysis

• For performing pixel analysis Canny edge detection algorithm is used [16]

• Helps in detection of fractal like objects



Fire Detector Module Pixel Analysis

• Output of pixel analysis is obtained by performing AND operation between frame analysis output and Canny edge detector output

Potential candidate fire pixels



Fire Detector Module Classifier

• This component further filters out the candidate pixels• Each pixel in obtained frame is compared to threshold value• The literature suggests this value should be greater than 180• The system uses three threshold values

1. Brightness Threshold

2. Canny Threshold

3. Flicker Threshold

Overview



Performance Evaluation


Demonstration



Metrics Precision Recall Accuracy Time elapsed to detect first positive frame

Static Parameters Flicker Threshold Canny Threshold Number of Frames Resolution of given video Frames per second for given video

Dynamic Parameters Brightness threshold



0 15 30 45 60 75 90 105 120 135 150 165 180 195 210 225 240 2550

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

TEST CASE 1

Precision

Recall

Accuracy

Bright Threshold

Prec

isio

n - R

ecal

l -Ac

cura

cy



0 15 30 45 60 75 90 105 120 135 150 165 180 195 210 225 240 2550

50000

100000

150000

200000

250000

300000

350000

TEST CASE 1

Time Taken (msec)

Bright Threshold

Tim

e Ta

ken

to d

etec

t fir

st p

osit

ive

fram

e (m

sec)

Overview



Conclusion


0 15 30 45 60 75 90 105 120 135 150 165 180 195 210 225 240 2550

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1Avg. Precision

Avg.Recall

Avg. Accuracy

Bright Threshold

Avg.

Pre

cisi

on -

Avg.

Reca

ll - A

vg. A

ccur

acy

Conclusion


System is showing average accuracy of 60% Precision and recall shows downward slope with increase

in threshold values The assumption of fire pixels having gray value more than

180 has proved wrong Equalization of histogram for each frame may prove

beneficial in improving the precision and recall of the system


Thank You

References


[1] Y. Do, “Flame detection in grey-scale images of a B/W camera”, in Sensor Review, vol. 34, no. 1, pp. 80–88, 2014.

[2] H. Yamagishi and J. Yamaguchi, “A contour fluctuation data processing method for fire flame detection using a color

camera”, in Industrial Electronics Society 26th Annual Confjerence of the IEEE, IEEE, 2000, vol. 2, pp. 824–829.

[3] L. Hongliang, L. Qing, and W. Sun’an, “A novel fire recognition algorithm based on flame’s Multi-features Fusion,”

in International Conference on Computer Communication and Informatics (ICCCI), 2012, pp. 1–6.

[4] F. Yuan, “An integrated fire detection and suppression system based on widely available video surveillance”, in Machine

Vision and Applications, 2010, vol. 21, no. 6, pp. 941–948.

[5] X. Zhou, F. Yu, Y. Wen, Z. Lu, and G. Song, “Early fire detection based on flame contours in video”, in Information

Technology Journal, 2010, vol. 9, no. 5, pp. 899–908.

[6] A. Çetin, K. Dimitropoulos, B. Gouverneur, N. Grammalidis, O. Günay, Y. Habiboglu, B. Töreyin, and S. Verstockt,

“Video fire detection–Review”, in Digital Signal Processing, 2013, vol. 23, no. 6, pp. 1827–1843.

[7] R. Bohush and N. Brouka, “Smoke and flame detection in video sequences based on static and dynamic features,”

in Signal Processing: Algorithms, Architectures, Arrangements, and Applications (SPA) , 2013, pp. 20–25.

[8] T. Song, B. Tang, M. Zhao, and L. Deng, “An accurate 3-D fire location method based on sub-pixel edge detection

and non-parametric stereo matching”, in Measurement, 2014.

References


[9] T. Li, M. Ye, F. Pang, H. Wang, and J. Ding, “An efficient fire detection method based on orientation feature” in International Journal of

Control, Automation and Systems, 2013, vol. 11, no. 5, pp. 1038–1045.

[10] L. Mengxin, X. Weijing, Z. Ying, and Z. Rui, “A new fire detection method based on integrated features,” in 24th Chinese Control and

Decision Conference (CCDC), 2012, pp. 965–970.

[11] M. Mueller, P. Karasev, I. Kolesov, and A. Tannenbaum, “Optical Flow Estimation for Flame Detection in Videos”, in IEEE

Transactions on Image Processing, 2013, vol. 22, no. 7, pp 2786 – 2797.

[12] P. Borges and E. Izquierdo, “A probabilistic approach for vision-based fire detection in videos”, in IEEE Transactions on Circuits and

Systems for Video Technology,IEEE, 2010 , vol. 20, no. 5, pp. 721–731.

[13] G. Marbach, M. Loepfe, and T. Brupbacher, “An image processing technique for fire detection in video images” in Fire safety

journal, 2006, vol. 41, no. 4, pp. 285–289.

[14] B. Ko, K. Cheong, and J. Nam, “Fire detection based on vision sensor and support vector machines” in Fire Safety

Journal, 2009, vol.44, no. 3, pp. 322–329.

[15] R. Lascio, A. Greco, A. Saggese and M. Vento, “Improving fire detection reliability by a combination of videoanalytics”, International

Conference on Image Analysis and Recognition (ICIAR), 2014.

[16] J. Canny, “A Computational Approach to Edge Detection”, in IEEE Transactions on Pattern Analysis and Machine Intelligence, IEEE,

1986, vol. 8, no. 6, pp. 679-698.

[17] The OpenCV Reference Manual, Revision 2.4.9.0, OpenCV Community, April 2014.

Research Project Presentation

Documents

Transcript of Research Project Presentation