How to separate overlapping objects?

asked 2016-02-18 17:42:21 -0600

Hi Developers, I want separate a binary image objects. To create the binary image i use Mog2 With ImageDepth Kinect, the result are objects (people) on the move, for the characteristics of the objects use FindCountours. The problem is that there are cases where objects overlap, I'm a beginner in Computer Vision. I use Opencv 3.0, Visual Studio 2015, Kinect Sdk 1.8, C++. the kinect camera is implemented top - down , objects(people) are head to the floor. In short I need to separate the overlapping objects. Greetings and sorry for my english

edit retag flag offensive close merge delete


This is a hard problem, but there are a couple of things you can do.

Use Canny Edge detector and erase any object pixels that lie on an edge. If there's a clear edge separating the objects, this should separate them. It is pretty likely that it will cause some false separations. Limiting this to known overlaps would help, if you can get a list.

Use one of the segmentation functions and seed it based on the centers of objects and split using that. This depends heavily on your seeding and algorithm parameters.

People tend to be different colors, so try running these in one of the color spaces that cleanly separates your typical examples.

Tetragramm gravatar imageTetragramm ( 2016-02-19 15:52:47 -0600 )edit

^^ don't post answers, then !

berak gravatar imageberak ( 2018-08-26 01:32:31 -0600 )edit

@ShaneP, we don't like your links, either ! (spam)

berak gravatar imageberak ( 2018-08-27 07:49:12 -0600 )edit