line extration in opencv_contrib

edit

Thank you!

I bought a tank model in a supermarket. Now I intend to detect gun barrel by LineSegmentDetector. But I have two issues. Firstly the edge lines of the tank and other uncorrelated lines are all detected. It is difficult to distinguish the edge lines of gun barrel. Secondly the detection result of LineSegmentDetector are not stright lines but many short lines. Could you give me some advice?

"detection result of LineSegmentDetector are not stright lines but many short lines" - yes, that's true. so, maybe it's the wrong tool for your job. if you still want to detect lines, have a look at HoughLines which will compute 'inifinite' lines, or HoughLinesP which will result in segments, too, but at least longer ones.

then, there's some doubt, if finding lines it the right idea at all. maybe findContours ?

Thanks for your help! The input of findContours is a binary image. Do you mean threshod the gray image? But the part of gun barrel may disappear in the process of threshoding.

The edge of gun barrel are two parallel lines. How about using HoughLines to detect lines of which slope are similar and distance satisfy a certain condition?

if your barrel has a clear contrast to the bg, threshold() will do fine. Sobel() and Canny() are alternative ways.

Could you tell me how to adjust the parameters to make the LineSegmentDetector output stright lines or shorter lines?

sample runs fine even on vs2008 here, what was the problem on your side ?

i did not write the lsd code, but i guess, the main purpose of it is to find very short pieces (that allow to get a gradient /orientation from this) . i don't see any way to make them longer. again, you're probably looking at the wrong tool. (yes, it's far easier to use, than HuoghLines, but again, very restricted)

After delete the Canny(image, image, 50, 200, 3); // Apply canny edge, I find the sample run well in VS13 but it cannot work well in VS12.

Thanks for your help!