0

My objective -

Input: A PNG floorplan (with many electrical equipment symbols on it), and a user who selects one of those symbols using a bounding box.

Output: The same PNG floorplan but with all matching symbols highlighted

I have been looking into feature detection as a way to find matching symbols, but I can't find any examples online of it used on 2D objects- I only ever see it used on photos or used live in videos. Does Feature Detection work for 2D objects as well? If not, why not?

For those interested, I have been developing in C#, using an OpenCV wrapper API called Emgu CV (it has all the OpenCV functions and some more).

Matthew Fishman
  • 23
  • 7

2 Answers2

1

You can take a look at research on logo recognition. You can use classical features detector such as sift or surf and then calculate from the extracted features some shape invariants like features triangle orientation.

Here is a classic paper to take a look for some ideas: Scalable Logo Recognition

Luca Del Tongo
  • 2,702
  • 16
  • 18
0

I guess your input is a binary / grayscale image with mainly lines, arrows, circles, ... Local Feature Detection (eg ORB, SURF, SIFT, ...) is best suited for "high entropy" images, I mean photos of scenes with a lot of texture. Here you have geometrical shapes, a geometrical method would be better. I think a (geometric) shape detection algorithm would be a better choice, such as the Generalized Hough Transform.

Vincent Vidal
  • 323
  • 2
  • 11