Object detector for unchanging object viewed from slightly different perstectives

asked 2016-02-10 20:03:09 -0500

I have seen how effective SURF and SIFT can be at feature matching 2d images, but I was wondering if training a cascade filter on a single object would be better for detecting a single object at different orientations and distances.

The goal would be to detect a raised landing pad-like object from a UAV when the object may be at the edge of a wide angle image, at any orientation, and/or from a substantial height. This object will be around a foot cubed and could have distinctive coloring and markings.


edit retag flag offensive close merge delete



maybe, you can just slap some markers, like aruco onto your landing platform ?

berak gravatar imageberak ( 2016-02-11 02:39:44 -0500 )edit

Like @berak said, use markers, technique has been proven and is succesfully used in several applications like this company.

StevenPuttemans gravatar imageStevenPuttemans ( 2016-02-11 08:28:18 -0500 )edit

Do those do well at a distance?

person_I_swear gravatar imageperson_I_swear ( 2016-02-11 10:39:27 -0500 )edit

Yes, they actually do. We even use large squared meter markers for drones flying relatively higher!

StevenPuttemans gravatar imageStevenPuttemans ( 2016-02-12 03:49:26 -0500 )edit