Bad performance with hough transform

[jkh13]

I am trying to code up a glyph detection process using XNA + AForge.NET and I have encountered some problems. At the moment I am using the following techniques:

1. Grayscale
2. Canny Edge Filter
3. Hough line transform

The problem is that the hough line transform is taking a very long time, much slower than my target 30 fps. I was wondering if there are any algorithms that detect lines in an image in realtime? or if I am using the hough line transform inefficiently?

[andrew.kirillov]

It is known that Hough line transformation is not very fast. You may read theory of the algorithm to have some understanding of it.

What type of glyphs are you trying to detect? I did a project some time ago about glyph recognition and augmented reality and it did no require lines detection at all.

[jkh13]

Hi Andrew, thanks for your reply.

I have realised that the hough transform may not be the best way to solve the problem. I have read your glyph detection algorithm and it is similar to what I am trying to achieve, the glyphs I am using are the same as yours.

My plan was to split the image into a grid and run a line detection algorithm such as hough on each grid, from there I can join the line segments obtained together to locate the quadrilateral. The line segments could even be extended so that the algorithm would be more robust to occlusion of the glyph. I am thinking a RANSAC algorithm might be more suitable and run much faster for the line detection.