I am working on a project to detect fire in CCTV or any other surveillance cameras. Up till now I have completed the following stages;
- Convert the RGB into HSV
- Sliced out the Value matrix from HSV
- Applied background subtraction model from openCV () to separate the foreground which is area of the fire.
- Applied morphological operations on the extracted foreground.
- Extracted contours from the foreground.
- ANDed the contour mask to original HSV
Now I am getting an HSV image where there are shades of fire color's in the Region of Interest (ROI)
I am getting the following output on the video where there is fire.
My problem is I am getting such output on videos where there is no fire but shades of red, yellow and orange are present with high intensity. How can I improve it further with eliminating false positives?