Object recognition by computer programs is the biggest promise of a endless range of new application and appliances (Internet of things) . Robotics, devices can be much more smart if they could be able to recognize objects viewed by a webcam.
So I experimented some software technics to reach this capability. In detail I concentrated into:
- finding points of interest in images and linking them by optimal triangles (Delaunay triangulation).
- Let users select which triangles best approximate the choosen object shape
- Using periferic points of selected triangles to find other instances of that object in different images or videos by mean of a “projection transform” (2D Homography) . I’ve still to do this but with main algorithm already made for this develop.
This picture is an example of first 2 steps:
If you wish directly experiment this function, below is the Flash app developed. You’ll have to:
- accept webcam input
- take image to elaborate pressing “Rileva” (Detect)
- after few seconds you’ll have the points of interest on right side and, with right click on image, select “Vedi Aree” (Show Areas)
Post a comment