Revision history [back]

The questions are related to the paper--"smoke detection algorithm for intelligent video surveillance system".

1 : How to iterate the region of blobs?

Assume that I have a 5 * 5 blob

(0, 0) (0, 1) (0, 2) (0, 3) (0, 4) (1, 0) (1, 1) (1, 2) (1, 3) (1, 4) (2, 0) (2, 1) (2, 2) (2, 3) (2, 4) (3, 0) (3, 1) (3, 2) (3, 3) (3, 4) (4, 0) (4, 1) (4, 2) (4, 3) (4, 4)

Then I would iterate the pixels with 2x2 blocks (1, 1) (1, 2) pre(1, 1) will compare with pixels cur(0, 0), cur(0, 1), cur(0, 2) (2, 1) (2, 2)

(1, 3) (1, 4) pre(1, 3) will compare with pixels cur(0, 2), cur(0, 3), cur(0, 4) (2, 3) (2, 4)

(3, 1) (3, 2) pre(3, 1) will compare with pixels cur(2, 0), cur(2, 1), cur(2, 2) (4, 1) (4, 2)

(3, 3) (3, 4) pre(3, 3) will compare with pixels cur(2, 2), cur(2, 3), cur(2, 4) (4, 3) (4, 4)

choose the largest values after compare(according to F) and save it into cs

Is this correct?

2 : What is the results after compare? should I save it as the value after compare(F), or save it as the vectors indicate the direction? I think it should be value, else it is hard to explain what is the meaning of

p = (cs / c) * 100%

3 : we should apply optical flow detection on the whole image or just apply it on the contours search by the previous step(connected component analysis)?

4 : the formula of Cw is a little bit weird, I think it should be

(I(t + 1) - B(t)) / B(t)

same as 3, apply Weber contrast on the whole image or on the contours search by the previous step(connected component analysis)?

I guess this should only apply on the candidate smoke searched by previous step(connected component).

Thanks a lot

The questions are related to the paper--"smoke detection algorithm for intelligent video surveillance system".

1 : How to iterate the region of blobs?

Assume that I have a 5 * 5 blob

(0, 0) (0, 1) (0, 2) (0, 3) (0, 4)
(1, 0) (1, 1) (1, 2) (1, 3) (1, 4)
(2, 0) (2, 1) (2, 2) (2, 3) (2, 4)
(3, 0) (3, 1) (3, 2) (3, 3) (3, 4)
(4, 0) (4, 1) (4, 2) (4, 3) (4, 4)
4)

The questions are related to the paper--"smoke detection algorithm for intelligent video surveillance system".

1 : How to iterate the region of blobs?

Assume that I have a 5 * 5 blob

(0, 0) (0, 1) (0, 2) (0, 3) (0, 4)
(1, 0) (1, 1) (1, 2) (1, 3) (1, 4)
(2, 0) (2, 1) (2, 2) (2, 3) (2, 4)
(3, 0) (3, 1) (3, 2) (3, 3) (3, 4)
(4, 0) (4, 1) (4, 2) (4, 3) (4, 4)

Then I would iterate the pixels with 2x2 blocks (1, 1) (1, 2) pre(1, 1) will compare with pixels cur(0, 0), cur(0, 1), cur(0, 2) (2, 1) (2, 2)

(1, 3) (1, 4) pre(1, 3) will compare with pixels cur(0, 2), cur(0, 3), cur(0, 4) (2, 3) (2, 4)

(3, 1) (3, 2) pre(3, 1) will compare with pixels cur(2, 0), cur(2, 1), cur(2, 2) (4, 1) (4, 2)

(3, 3) (3, 4) pre(3, 3) will compare with pixels cur(2, 2), cur(2, 3), cur(2, 4) (4, 3) (4, 4)

choose the largest values after compare(according to F) and save it into cs

Is this correct?

2 : What is the results after compare? should I save it as the value after compare(F), or save it as the vectors indicate the direction? I think it should be value, else it is hard to explain what is the meaning of

p = (cs / c) * 100%

3 : we should apply optical flow detection on the whole image or just apply it on the contours search by the previous step(connected component analysis)?

4 : the formula of Cw is a little bit weird, I think it should be

(I(t + 1) - B(t)) / B(t)

B(t) / (number of pixels)

but the paper mention it is F(t+1)? same as 3, apply Weber contrast on the whole image or on the contours search by the previous step(connected component analysis)?

I guess this should only apply on the candidate smoke searched by previous step(connected component).

Thanks a lot

The questions are related to the paper--"smoke detection algorithm for intelligent video surveillance system".

1 : How to iterate the region of blobs?

Assume that I have a 5 * 5 blob

(0, 0) (0, 1) (0, 2) (0, 3) (0, 4)
(1, 0) (1, 1) (1, 2) (1, 3) (1, 4)
(2, 0) (2, 1) (2, 2) (2, 3) (2, 4)
(3, 0) (3, 1) (3, 2) (3, 3) (3, 4)
(4, 0) (4, 1) (4, 2) (4, 3) (4, 4)

Then I would iterate the pixels with 2x2 blocks (1, 1) (1, 2) pre(1, 1) will compare with pixels cur(0, 0), cur(0, 1), cur(0, 2) (2, 1) (2, 2)

(1, 3) (1, 4) pre(1, 3) will compare with pixels cur(0, 2), cur(0, 3), cur(0, 4) (2, 3) (2, 4)

(3, 1) (3, 2) pre(3, 1) will compare with pixels cur(2, 0), cur(2, 1), cur(2, 2) (4, 1) (4, 2)

(3, 3) (3, 4) pre(3, 3) will compare with pixels cur(2, 2), cur(2, 3), cur(2, 4) (4, 3) (4, 4)

choose the largest values after compare(according to F) and save it into cs

Is this correct?

2 : What is the results after compare? should I save it as the value after compare(F), or save it as the vectors indicate the direction? I think it should be value, else it is hard to explain what is the meaning of

p = (cs / c) * 100%

3 : we should apply optical flow detection on the whole image or just apply it on the contours search by the previous step(connected component analysis)?

4 : the formula of Cw is a little bit weird, I think it should be

(I(t + 1) - B(t)) / B(t) / (number of pixels)

but the paper mention it is F(t+1)? same as 3, apply Weber contrast on the whole image or on the contours search by the previous step(connected component analysis)?

I guess this should only apply on the candidate smoke searched by previous step(connected component).

Thanks a lot

Edit : 1: I tried it with my guess, the result seem reasonable Find reasonable.Find the similarity on whole image could get better result result, don't find it on a single regionregion.

Edit 2 : Some videos about the result

• burnOutContest
• explosion
• fumogeno1
• fumegeno2

The results are far from perfect,I don't know the reason yet. Is it because I use bilinear to down sample the image and using gaussian kernel to remove noise instead of discrete wavelet transform? Misunderstanding of the algorithm?Wrong parameters?

If you want to develop a reliable real time smoke detection algorithm which could run on low cost devices($100~$200), which paper would you recommend? Which paper would you

The questions are related to the paper--"smoke detection algorithm for intelligent video surveillance system".

1 : How to iterate the region of blobs?

Assume that I have a 5 * 5 blob

(0, 0) (0, 1) (0, 2) (0, 3) (0, 4)
(1, 0) (1, 1) (1, 2) (1, 3) (1, 4)
(2, 0) (2, 1) (2, 2) (2, 3) (2, 4)
(3, 0) (3, 1) (3, 2) (3, 3) (3, 4)
(4, 0) (4, 1) (4, 2) (4, 3) (4, 4)

Then I would iterate the pixels with 2x2 blocks (1, 1) (1, 2) pre(1, 1) will compare with pixels cur(0, 0), cur(0, 1), cur(0, 2) (2, 1) (2, 2)

(1, 3) (1, 4) pre(1, 3) will compare with pixels cur(0, 2), cur(0, 3), cur(0, 4) (2, 3) (2, 4)

(3, 1) (3, 2) pre(3, 1) will compare with pixels cur(2, 0), cur(2, 1), cur(2, 2) (4, 1) (4, 2)

(3, 3) (3, 4) pre(3, 3) will compare with pixels cur(2, 2), cur(2, 3), cur(2, 4) (4, 3) (4, 4)

choose the largest values after compare(according to F) and save it into cs

Is this correct?

2 : What is the results after compare? should I save it as the value after compare(F), or save it as the vectors indicate the direction? I think it should be value, else it is hard to explain what is the meaning of

p = (cs / c) * 100%

3 : we should apply optical flow detection on the whole image or just apply it on the contours search by the previous step(connected component analysis)?

4 : the formula of Cw is a little bit weird, I think it should be

(I(t + 1) - B(t)) / B(t) / (number of pixels)

but the paper mention it is F(t+1)? same as 3, apply Weber contrast on the whole image or on the contours search by the previous step(connected component analysis)?

I guess this should only apply on the candidate smoke searched by previous step(connected component).

Thanks a lot

Edit 1: I tried it with my guess, the result seem reasonable.Find the similarity on whole image could get better result, don't find it on a single region.

Edit 2 : Some videos about the result

• burnOutContest
• explosion
• fumogeno1
• fumegeno2

The results are far from perfect,I don't know the reason yet. Is it because I use bilinear to down sample the image and using gaussian kernel to remove noise instead of discrete wavelet transform? Misunderstanding of the algorithm?Wrong parameters?

If you want to develop a reliable real time smoke detection algorithm which could run on low cost devices($100~$200), which paper would you recommend? Which paper would you recommend?

The questions are related to the paper--"smoke detection algorithm for intelligent video surveillance system".

1 : How to iterate the region of blobs?

Assume that I have a 5 * 5 blob

(0, 0) (0, 1) (0, 2) (0, 3) (0, 4)
(1, 0) (1, 1) (1, 2) (1, 3) (1, 4)
(2, 0) (2, 1) (2, 2) (2, 3) (2, 4)
(3, 0) (3, 1) (3, 2) (3, 3) (3, 4)
(4, 0) (4, 1) (4, 2) (4, 3) (4, 4)

Then I would iterate the pixels with 2x2 blocks (1, 1) (1, 2) pre(1, 1) will compare with pixels cur(0, 0), cur(0, 1), cur(0, 2) (2, 1) (2, 2)

(1, 3) (1, 4) pre(1, 3) will compare with pixels cur(0, 2), cur(0, 3), cur(0, 4) (2, 3) (2, 4)

(3, 1) (3, 2) pre(3, 1) will compare with pixels cur(2, 0), cur(2, 1), cur(2, 2) (4, 1) (4, 2)

(3, 3) (3, 4) pre(3, 3) will compare with pixels cur(2, 2), cur(2, 3), cur(2, 4) (4, 3) (4, 4)

choose the largest values after compare(according to F) and save it into cs

Is this correct?

2 : What is the results after compare? should I save it as the value after compare(F), or save it as the vectors indicate the direction? I think it should be value, else it is hard to explain what is the meaning of

p = (cs / c) * 100%

3 : we should apply optical flow detection on the whole image or just apply it on the contours search by the previous step(connected component analysis)?

4 : the formula of Cw is a little bit weird, I think it should be

(I(t + 1) - B(t)) / B(t) / (number of pixels)

but the paper mention it is F(t+1)? same as 3, apply Weber contrast on the whole image or on the contours search by the previous step(connected component analysis)?

I guess this should only apply on the candidate smoke searched by previous step(connected component).

Thanks a lot

Edit 1: I tried it with my guess, the result seem reasonable.Find the similarity on whole image could get better result, don't find it on a single region.

Edit 2 : Some videos about the result

• burnOutContest
• explosion
• fumogeno1
• fumegeno2
• fumegeno3
• fumegeno4
• fumegeno5
• fumegeno6

The results are far from perfect,I don't know the reason yet. Is it because I use bilinear to down sample the image and using gaussian kernel to remove noise instead of discrete wavelet transform? Misunderstanding of the algorithm?Wrong parameters?

If you want to develop a reliable real time smoke detection algorithm which could run on low cost devices($100~$200), which paper would you recommend?

The questions are related to the paper--"smoke detection algorithm for intelligent video surveillance system".

1 : How to iterate the region of blobs?

Assume that I have a 5 * 5 blob

(0, 0) (0, 1) (0, 2) (0, 3) (0, 4)
(1, 0) (1, 1) (1, 2) (1, 3) (1, 4)
(2, 0) (2, 1) (2, 2) (2, 3) (2, 4)
(3, 0) (3, 1) (3, 2) (3, 3) (3, 4)
(4, 0) (4, 1) (4, 2) (4, 3) (4, 4)

Then I would iterate the pixels with 2x2 blocks (1, 1) (1, 2) pre(1, 1) will compare with pixels cur(0, 0), cur(0, 1), cur(0, 2) (2, 1) (2, 2)

(1, 3) (1, 4) pre(1, 3) will compare with pixels cur(0, 2), cur(0, 3), cur(0, 4) (2, 3) (2, 4)

(3, 1) (3, 2) pre(3, 1) will compare with pixels cur(2, 0), cur(2, 1), cur(2, 2) (4, 1) (4, 2)

(3, 3) (3, 4) pre(3, 3) will compare with pixels cur(2, 2), cur(2, 3), cur(2, 4) (4, 3) (4, 4)

choose the largest values after compare(according to F) and save it into cs

Is this correct?

2 : What is the results after compare? should I save it as the value after compare(F), or save it as the vectors indicate the direction? I think it should be value, else it is hard to explain what is the meaning of

p = (cs / c) * 100%

3 : we should apply optical flow detection on the whole image or just apply it on the contours search by the previous step(connected component analysis)?

4 : the formula of Cw is a little bit weird, I think it should be

(I(t + 1) - B(t)) / B(t) / (number of pixels)

but the paper mention it is F(t+1)? same as 3, apply Weber contrast on the whole image or on the contours search by the previous step(connected component analysis)?

I guess this should only apply on the candidate smoke searched by previous step(connected component).

Thanks a lot

Edit 1: I tried it with my guess, the result seem reasonable.Find the similarity on whole image could get better result, don't find it on a single region.

Edit 2 : Some videos about the result

• burnOutContest
• explosion
• fumogeno1
• fumegeno2
• fumegeno3
• fumegeno4
• fumegeno5
• fumegeno6

The results are far from perfect,I don't know the reason yet. Is it because I use bilinear to down sample the image and using gaussian kernel to remove noise instead of discrete wavelet transform? Misunderstanding of the algorithm?Wrong parameters?

If you want to develop a reliable real time smoke detection algorithm which could run on low cost devices($100~$200), which paper would you recommend?