Object recognition software in 3D images

Authors: ROBERT CUPEC, IVAN VIDOVIĆ, DAMIR FILKO, PETRA PEJIĆ

Faculty: Fakultet elektrotehnike, računarstva i informacijskih tehnologija Osijek

Country: Croatia

e-mail: robert.cupec@ferit.hr

My innovation is a Software that identifies one or more target objects in images captured with 3D cameras such as Kinect, ASUS Xtion Pro Live, Orbbec Astra, Intel RealSense, and more.  The software needs to build a 3D model of the target object into a triangular mesh. The core of the software is an algorithm consisting of three main steps: (1) Detect flat and convex surfaces in 3D images; (2) Add instances of convex templates (CTI) descriptors for each convex surface; (3) Generate hypotheses about the existence of target objects at specific locations in the picture; (4) Evaluate hypotheses and reject unlikely hypotheses.  Hypotheses remaining after the described procedure represent the final result – the interpretation of the scene, which consists of a series of data structures.

It recognises objects placed on a flat surface in any position, and when there are multiple objects on the stage, some of which partially occlude the target objects.