Detecting Thick Line Intersections
Kevin Marsh from United States  [2 posts]
8 year
New to RoboRealm! Looks like a great tool so far.

Any suggestions to detect the points that roads intersect in an image of a scanned map like https://i.imgur.com/xWI4dem.jpg ?

Steven Gentner from United States  [1446 posts] 8 year
Not yet your final results, but the attached robofile (jpg result included too) can be used to find the actual roads (green color overlaid on original image). The thought was to first identify roads which can then lead to the detection of intersections though dilation and erosion after the roads are detected.

The question is which intersections are considered valid as there are quite a few suspect intersections ... i.e. which intersections need to be identified or not.



Kevin Marsh from United States  [1 posts] 8 year
That's pretty awesome!

Not sure if spurious intersections is that big of a problem, we'll be using a pattern matching algorithm that should tolerate that. However, less is more. If we can somehow just detect a percentage based on the thickest lines with a high confidence that'd be preferable to trying to get every one 100% correct.

FYI The end game here is to match the scan up with a list of known intersections from GIS (X/Ys in longitude/latitude are attached for this particular county) by determining the proper translation.

The attached image shows the road network and the extracted intersection dots, this is what we're trying to match.

Speaking of which, does RoboRealm do this sort of pattern matching?

Thanks again!


Steven Gentner from United States  [273 posts] 8 year

RoboRealm does have that sort of pattern matching but not in a way that is useful for you. Its a common technique for object recognition where keypoints (i.e. your intersections) are matched to a know database of keypoints prior to final comparison. So while this is present in the Object Recognition module its not in a form that would work for you.

What you are looking at is a Ransac type method that can tolerance outliers. I wonder if you would find more useful research done in star matching since its not unlike matching the points of intersections AND would work on a very large database ... something the OR module would NOT. Because of the database size that you would be working with (assuming this since you mentioned GIS) the technique would have to be extremely fast and built to filter out false matches very quickly.

Its a great project ... but will require quite a bit of software dev to get that to work reliably.

Kevin Marsh from United States  [2 posts] 8 year
Quick followup, I finally got around to testing this out.

The line detection works great, what's the next step in terms of detecting intersections of the lines and outputting as X, Y? I think as a starting point only 2 that intersect (e.g., a + pattern) should be detected.


This forum thread has been closed due to inactivity (more than 4 months) or number of replies (more than 50 messages). Please start a New Post and enter a new forum thread with the appropriate title.

 New Post   Forum Index