OpenCV: Fire detection and cutting

1 modules

import cv2
import numpy as np

2 Conditions

• R>redThre: In a fire image, R is always higher than average value
• R>=G>=B
• S>=0.2
• S >= (255 - R)*saturationTh/redThre

where:

Then, the point which satisfy above conditions will be white and other points will be black.

In addition, in some bright point( such as the center in the following picture), R may be smaller than G. I add another condition: R>250 and G>250 will be recognized as fire.

cap = cv2.VideoCapture("3.flv")

redThre = 105
saturationTh = 42
const = 240

while cap.isOpened():
    _, img = cap.read()
    img = cv2.pyrDown(img)
    B = img[:, :, 0]
    G = img[:, :, 1]
    R = img[:, :, 2]

    #找到B、G、R中的最小值
    minValue = np.array(np.where(R <= G, np.where(G <= B, R, np.where(R <= B, R, B)), np.where(G <= B, G, B)))
    total = R+B+G
    total_1 = total[:, :]+1# to avoid R+B+G=0

    #计算S分量
    m_divide_t = np.divide(minValue, total_1)
    S = 1 - 3*m_divide_t
   
    #需满足条件：1 R>redThre 2 R>=G>=B; 3 S>=0.2; 4 S >= (255 - R)*saturationTh/redThre
#并且，当R和G都大于250时，也识别为火焰，为了避免火焰完全呈白色的情况   
    fireImg = np.array(np.where(R > redThre, np.where(R >= G, np.where(G >= B, np.where(S >= 0.2, np.where(S >= (255 - R)*saturationTh/redThre, 255, 0), 0), 0), 0), 0))
    fireImg_bu = np.array(np.where(R > 250, np.where(G > 250, 255, 0), 0))

3 merge the images

Merge the fireImg and fireImg _bu. And use GaussianBlur and threshold to optimize the image.

gray_fireImg1 and gray_fireImg2 are used to contrast influence of morphologyEx.

#将找到的火焰点都结合起来，并用高斯模糊和阈值处理图像
gray_fireImg = np.zeros([fireImg.shape[0], fireImg.shape[1], 1], np.uint8)
gray_fireImg[:, :, 0] = fireImg + fireImg_bu
gray_fireImg = cv2.GaussianBlur(gray_fireImg, (7, 7), 0)
_, gray_fireImg1 = cv2.threshold(gray_fireImg, 10, 255, cv2.THRESH_BINARY)

gray_fireImg2 = cv2.morphologyEx(gray_fireImg2, cv2.MORPH_CLOSE, kernel)

4 show the image

Show the image, and read the next image when handling a video.

    cv2.imshow("0", img)
    cv2.imshow("1", gray_fireImg1)
    cv2.imshow("2", gray_fireImg2)
    _, img = cap.read()

    if cv2.waitKey(20) == 27:
        break

cv2.destroyAllWindows()

5 result

Left one is gray_fireImg1.
Middle one is gray_fireImg2.
Right one is source video.

It seems that there is little difference between gray_fireImg1 and gray_fireImg2.

6 the whole code

import cv2
import numpy as np

cap = cv2.VideoCapture("3.flv")

redThre = 105
saturationTh = 42
const = 240

while cap.isOpened():
    _, img = cap.read()
    img = cv2.pyrDown(img)
    B = img[:, :, 0]
    G = img[:, :, 1]
    R = img[:, :, 2]

    #找到B、G、R中的最小值
    minValue = np.array(np.where(R <= G, np.where(G <= B, R, np.where(R <= B, R, B)), np.where(G <= B, G, B)))
    total = R+B+G
    total_1 = total[:, :]+1


    m_divide_t = np.divide(minValue, total_1)

    #计算S分量
    S = 1 - 3*m_divide_t

    #需满足条件：0 R>redThre 1 R>=G>=B; 2 S>=0.2; 3 S >= (255 - R)*saturationTh/redThre
    #并且，当R和G都大于250时，也识别为火焰，为了避免火焰完全呈白色的情况
    fireImg = np.array(np.where(R > redThre, np.where(R >= G, np.where(G >= B, np.where(S >= 0.2, np.where(S >= (255 - R)*saturationTh/redThre, 255, 0), 0), 0), 0), 0))
    fireImg_bu = np.array(np.where(R > 250, np.where(G > 250, 255, 0), 0))

    gray_fireImg = np.zeros([fireImg.shape[0], fireImg.shape[1], 1], np.uint8)
    gray_fireImg[:, :, 0] = fireImg + fireImg_bu
    gray_fireImg = cv2.GaussianBlur(gray_fireImg, (7, 7), 0)
    _, gray_fireImg1 = cv2.threshold(gray_fireImg, 10, 255, cv2.THRESH_BINARY)

    gray_fireImg2 = cv2.morphologyEx(gray_fireImg2, cv2.MORPH_CLOSE, kernel)

    cv2.imshow("0", img)
    cv2.imshow("1", gray_fireImg1)
    cv2.imshow("2", gray_fireImg2)
    _, img = cap.read()

    if cv2.waitKey(20) == 27:
        break

cv2.destroyAllWindows()