ZED stereo object recognition

asked 2017-04-24 13:20:11 -0600

Alexia gravatar image

updated 2017-04-25 01:20:17 -0600

We're working on object recognition and approach, outdoors. The problem we have is, we're not sure what the right way to do object-recognition with stereo camera is. We were trying to do with features but the object was rarely recognised, thus we were thinking of doing it by colour as the object has a very distinct colour (red). We need to both recognise the object and calculate the distance to the object using our stereo camera (ZED).

TL;DR

What is the proper way to recognise and calculate the distance to an object using a stereo camera?

Best regards Alexia

edit retag flag offensive close merge delete

Comments

1

would you care to explain, how this is related to opencv ?

berak gravatar imageberak ( 2017-04-24 13:21:55 -0600 )edit

OpenCV is an image-processing library and we're trying to do image-processing to recognise and object and such.

Alexia gravatar imageAlexia ( 2017-04-25 01:22:47 -0600 )edit

you use example given in opencv 1 calibration.cpp 2 stereo_calib.cpp 3 stereo_match.cpp

LBerger gravatar imageLBerger ( 2017-04-25 01:51:27 -0600 )edit

Two distinct problems:

  • object detection / recognition ==> Deep Learning, Feature detection/Bag of Words, Cascade of Classifier, HOG, etc.
  • get object distance

If you are using the ZED stereo cameras, you should be able to get directly the depth map with the ZED SDK no? Otherwise if you have to do the stereo reconstruction manually (in general not in the case of the ZED cameras) see the comment above.

Eduardo gravatar imageEduardo ( 2017-04-25 04:20:53 -0600 )edit