sabra's profile - activity

@blacksailer & @afanasjev i did all things but in build step have error about find libs and gst/gst.h can you give me some advice?

hsv[1]/255 <0.2 ? i this right for smoke?

i have a program that check values of pixels to be in range

but the result surprize me!

only just part of image affected and smoky area arnt in range!

please help me

#include <opencv2/core/core.hpp>
#include <opencv2/imgproc/imgproc.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/video/video.hpp>


using namespace std;
using namespace cv;

void colorDetect(Mat mFrame,int redThreshold, double saturationThreshold)
{
    Mat temp;
    GaussianBlur(mFrame, temp, Size(3, 3), 0);
    Mat img;
    cvtColor(temp,img,CV_BGR2HSV);


    for (int i = 0; i < temp.rows; i++) {
        for (int j = 0; j < temp.cols; j++) {


            Vec3b& hsv = img.at<Vec3b>(i, j);

            Vec3b& v = temp.at<Vec3b>(i, j);

                    if((abs(v[2]-v[1])<=20 && abs(v[0]-v[1])<=20 && abs(v[2]-v[0])<=20 && hsv[1]<=0.4)  ){
                        mFrame.at<uchar>(i, j) = 255;

                    }
                    else{

                        mFrame.at<uchar>(i, j) = 0;
                    }


        }
    }

    imshow("mframe",mFrame);
    cvWaitKey(0);
}

int main(){

    Mat frame=imread("D:/121.jpg",1);
    colorDetect(frame,150,0.12);


    return 1;
}

hsv is better or rgb or ycrcb? im using: colorDetect(RGB based)+svm.predict

if i add red moving obj to negative,true alarms reduced

@Tetragramm @StevenPuttemans

the subject is clear friends, i can detect fire very good by svm, but when i add negative sample about moving red obj, again in the test step , a moving red object detected as fire!!

i dont know actualy ,this is part of an open source project,i adjusted it but its slow @LBerger

i edited post, your solution was good , but how about density? tnx for guide me

@LBerger

i have amethods like below that have good results:

void TargetExtractor::denoise(int ksize, int threshold)
{
    int r = (ksize - 1) / 2;
    if (r <= 0) {
        return;
    }

    Mat density;
    calcDensity(mMask, density, ksize);

    for (int i = r; i < mMask.rows - r; i++) {
        for (int j = r; j < mMask.cols - r; j++) {
            int count = density.at<int>(i, j);
            if (count < threshold) {
                mMask.at<uchar>(i, j) = 0;
            }
        }
    }
}

But its time is 120-140 ms!!! can you suggest me another methods?

edit 2: tnx to reply guys please notice that the:

in the code most time related to calcdensity:

/ this function is an alternative to the old one which is implemented with 'sum'
void calcDensity(const Mat& mask, Mat& density, int ksize)
{
    int r = (ksize - 1) / 2;
    if (r <= 0) {
        return;
    }

    density = Mat::zeros(mask.size(), CV_32SC1);

    int rowBound = density.rows - r, colBound = density.cols - r;

    Rect rect(0, 0, ksize, ksize);
    density.at<int>(r, r) = cvRound(sum(mask(rect))[0] / 255);

    for (int j = r + 1; j < colBound; j++) {
        int col1 = j - r - 1, col2 = j + r;
        int delta = 0;
        for (int k = 0; k < ksize; k++) {
            delta += mask.at<uchar>(k, col2) - mask.at<uchar>(k, col1);
        }
        density.at<int>(r, j) = density.at<int>(r, j - 1) + delta / 255;
    }

    for (int i = r + 1; i < rowBound; i++) {
        int row1 = i - r - 1, row2 = i + r;
        int delta = 0;
        for (int k = 0; k < ksize; k++) {
            delta += mask.at<uchar>(row2, k) - mask.at<uchar>(row1, k);
        }
        density.at<int>(i, r) = density.at<int>(i - 1, r) + delta / 255;
    }

    for (int i = r + 1; i < rowBound; i++) {
        for (int j = r + 1; j < colBound; j++) {
            int delta = (mask.at<uchar>(i + r, j + r) - mask.at<uchar>(i - r - 1, j + r) -
                mask.at<uchar>(i + r, j - r - 1) + mask.at<uchar>(i - r - 1, j - r - 1)) / 255;
            density.at<int>(i, j) = density.at<int>(i - 1, j) + density.at<int>(i, j - 1) -
                density.at<int>(i - 1, j - 1) + delta;
        }
    }
}

svm is better than colorspace methods!

@berak @StevenPuttemans

this is what i wanted

check your camera with vlc,if vlc cant open it,its a problem with your camera change your camera, else put your code here

i searched sir, its all about "methods" not "result of methods" @berak

also check channel of image

first add include file of opencv -- in my system: C:\opencv\build\include

then add lib directory -- in my system: C:\opencv\build\x86\vc11\lib --->note : im using vs2012

then add names of libs to input linker:

note: "im using opencv 2.4.9" what about u?

opencv_calib3d249d.lib opencv_contrib249d.lib opencv_core249d.lib opencv_features2d249d.lib opencv_flann249d.lib opencv_gpu249d.lib opencv_highgui249d.lib opencv_imgnd249d.lib opencv_imgproc249d.lib opencv_ml249d.lib opencv_nonfree249d.lib opencv_objdetect249d.lib opencv_ocl249d.lib opencv_photo249d.lib opencv_stasm249d.lib opencv_stitching249d.lib opencv_superres249d.lib opencv_ts249d.lib opencv_video249d.lib opencv_videostab249d.lib

if you are using v2013 you must add v12 of lib

for another example

if you are using v2012 you must add v11 of lib

best methods for fire detection using image processing? i saw svm,and color based (hsv or ycrcb) methods, i don't have 2000 negative and positive sample to train svm and see its result, so i hope someone guide me that this method is ok or not? note that im not a researcher,im just working on open source codes
best wishes

how make dst uchar? what changes are need? sorry im inexperienced @berak

so? what can i do? please @berak

i cheked like below

        Mat lookup(1,256,CV_32S, uniform); // make a Mat from it
        bool x=lookup.isContinuous();
        int cnn=lookup.channels();
        int y=lookup.total();
        int z= lbp_image.depth();
            LUT(lbp_image,lookup,lbp_image);// =====> error appear here

        Mat p = spatial_histogram(
                lbp_image, /* lbp_image */
                static_cast<int>(59), /* number of possible patterns */
                _grid_x, /* grid size x */
                _grid_y, /* grid size y */
                true);

x--> true

cnn-->1

y-->256

z-->4

@berak

@berak i did all but : assertion failed : (lutcn == cn || lutcn == 1) && lut.total() == 256 && lut.isContinuous() && (src.depth() == CV_8U || src.depth() == CV_8S) in function LUT

tnx alot but in second link what is your mean?

 static_cast<int>(std::pow(2.0, static_cast<double>(_neighbors))),

this result is 2^8 so

should i set it to static_cast<int>(59) ?

@berak : i edited it,if you need more details, i must first sort my code :)

hi, im using opencv func for uniform lbp (LUT ) and accuracy of recognition get worth but i read that this will not happen,what am i wrong?

read images

resize each image

lut(images,lookup table,images);

train images detection

@berak i changed my code to your suggestion,my trained model reduced from 6,7 Mb to 6,3 and accuracy is scandal!!!! why?