Geek Answers Handbook

Area of ​​Interest Uniqueness and Identity

I am currently working on a presentation of computer vision with OpenCV. The app includes goal identification and feature identification. Actually, I will have a target cross in the visible area and slowly go through it after a couple of seconds. This should give me 50-60 frames from the camera in which I can find the target.

We have successfully implemented detection algorithms using SWT and OCR (all targets have alphanumeric identifiers, which makes them relatively easy to distinguish). I want to use as much data as possible from all 50-60 snapshots of each target. To do this, I need to somehow determine that a specific ROI of image 2 contains the same goal as another ROI from image 1.

That I ask for a little advice from someone who may have come across this before. How can I easily / quickly determine, with a reasonable margin of error, that ROI # 2 has the same purpose as ROI # 1? My first instinct looks something like this:

  • Target detection in frame 1.
  • Calculate the specific unique functions of each of the targets in frame 1. Save.
  • Get frame 2.
  • Immediately find ROIs that have the same features as in step 2. Grab them and send them down the line for further processing, skipping step 5.
  • Detecting new targets in frame 2.
  • Transferring targets to the stream to calculate the shape, color, GPS coordinates, etc.
  • Pour, rinse, repeat.

I think that the capabilities of SURF or SIFT may be a way to accomplish this, but I am worried that they might have trouble identifying targets as the same from frame to frame due to color distortion or fading. I do not know how to set the threshold for SIFT / SURF functions.

Thank you for any light you can shed on this issue.

+4
source share
1 answer

One thing you can do is localize brightness levels and possibly saturation. If you are not using extended space such as YCrCb or HSV, I suggest you try them.

Can you assume that the object is not moving too fast? If you load the previous position into the discovery procedure, you can reduce the size of the window you are looking at. The same thing happens with speed and direction of movement.

I have successfully used the histogram and described the shape of the area, to reliably detect it, you can use it or add it to the SURF / SIFT classifier.

+2
source

More articles:


All Articles