erogol2's profile - activity

How could I add GTK support to my compiled Opencv 3.0 without compiling from scratch?

I want to apply the same operation of cvConvertScale directly to Mat without changing into IplImage. How should I do the same to Mat structure?

What is the most efficient way to add scalar to a particular channel of an opencv image?

I give double vectors with following example values; (don't think about "Blas1")

Blas1 2.92631e-18 
Blas1 3.98388e-20 
Blas1 9.31403e-19 
Blas1 6.41224e-19 
Blas1 8.68721e-24 
Blas1 1.07387e-21 
Blas1 5.82306e-21 
Blas1 2.36567e-18 
Blas1 5.5498e-19 
Blas1 2.93748e-19 
Blas1 2.54335e-18 
Blas1 1.92756e-23 
Blas1 5.49794e-20 
Blas1 4.14459e-20 
Blas1 8.69651e-20 
Blas1 2.97606e-21 
Blas1 2.81848e-23 
Blas1 2.37574e-21 
Blas1 2.49897e-19 
Blas1 3.69965e-20 
Blas1 8.56015e-20 
Blas1 7.00554e-20 
Blas1 2.38473e-18 
Blas1 1.19531e-24 
Blas1 4.78057e-19 
Blas1 3.69334e-20 
Blas1 5.75832e-20 
Blas1 1.46834e-21 
Blas1 1.33062e-19 
Blas1 1.41787e-19 
Blas1 5.04448e-20 
Blas1 1.94257e-18 
Blas1 2.97708e-19 
Blas1 6.41224e-19 
Blas1 1.16371e-23 
Blas1 3.79272e-20 
Blas1 4.31626e-23 
Blas1 3.14873e-18 
Blas1 3.23783e-23 
Blas1 8.00364e-26 
Blas1 1.08844e-19 
Blas1 9.714e-18 
Blas1 4.06371e-20 
Blas1 1.866e-19 
Blas1 5.05251e-25 
Blas1 2.39448e-21 
Blas1 9.57913e-19 
Blas1 4.84601e-22

However, after the normalization step they are all casted to 0. All of the values are kept as double. Is it a bug or impossible expectation to process such small numbers?

I try to install opencv3 with python2 support. I give all the necessary cmake flags with the correst set of paths but still when I cmake it raises following error.

File "<string>", line 1, in <module>
ImportError: No module named 'numpy'

Ans if I build it with this configuration, I cannot import opencv from python as well. Here is my cmake flags for python

What I miss for the correct compilation of Opencv with python support?

I try to convert a Mat image object to float array which includes the data in row major order. Which means 1D array includes the first row of the image, then the second row etc. At the end, it repeats this for the next image channel.

Here is the silly pseudo description float *A = [first_row_first_channel, second_row_first_channel, ..., first_row_second_channel, ...]

Does Mat img.data provides this ordering or it gives something different? If this is different how can I get the desired orientation ?

I try to experiment with HOG multi scale detector but cannot understand the exact behavior of the scale factor parameter. Could you explain what is the intuition behind this parameter?

I try to run a pedestrian detector with hog.detectMultiScale(). However, I realized that if the given image is smaller than a threshold, it exits the program without any warning. I also did not see any comment in the docs about any limitation of image size. I am using OpenCV3 and here is the section of my code with a naive solution.;

                Rect bb = boundingBoxes[i];

                // ignore too small bbs
                if(bb.area() < BB_THRESHOLD || bb.height < 60 || bb.width < 60)
                    continue;


                cout << bb.height << endl;
                cout << bb.width << endl;

                // get bb image and run detector
                Mat bb_image = fg(bb);
                hog.detectMultiScale(bb_image, found, 0, Size(5,5), Size(30,30), 1.0, 0, true);

However if I make smaller Size(30,30) to Size(10,10), I also need to increase the check sizes. I could not figure out what is the exact size threshold but this is the case. Any idea for the solution or does it seem like a bug in opencv?

Thanks for the return. Which tool gives me the connecteComponents. Do you mean finding contours?

I used BackgroundSubtractorMOG2 for foreground detection and obtain a mask to apply frames before pedestrian detection by HOG. However, I could not see any argument in the detect() method of HOG. How can I use this mask to reduce the computation time of the detection process?

I far as I see detect() only returns a point whereas detectMultiScale() returns a detection bounding box. Is there any way to replicate this for detect() function as well?

@berak there is no calibration for PTZ but the static camera. I try to click on a place on the view of that static camera and move the PTZ to that place by a correct level of tilt and zoom level.

I have 2 cameras. One a simple network camera and a PTZ camera. The simple one is statically located and constantly taking pictures of the same view. I screen the captured view by the simple camera. I try to zoom PTZ to a place I clicked on that screen. There are different challenges for this problem. The most catchy 2 of them are;

1. PTZ should know, what level of zoom is required for the clicked place. Too much zoom might lose the view and too low be to coarse.
2. PTZ should know how much and in what direction it should move to take the region of interest in its scope.

For these two challenges what should I do? Do you have any suggestions, references or pointers in mind? Any comment is welcome.

Addition1 : As far as I learned from WEB, this problem is called slew to cue.

I attempt to develope a OpenCV app based on AXIS PTZ camera. It has a SDK but I could not really find out a way to get frames into OpenCV matrices. If there are some experienced ones, could you tell me the best practice to use AXISPTZ with OpenCV?

Basic expectation is to direct PTZ camera into the motion on the scene to be more precise about the project. For instance it will direct to the guy walking on a parking lot.