Authors:
Nicola Greggio
1
;
José Gaspar
2
;
Alexandre Bernardino
2
and
José Santos-Victor
2
Affiliations:
1
ARTS Lab - Scuola Superiore S. Anna and Instituto Superior Técnico, Italy
;
2
Instituto Superior Técnico, Portugal
Keyword(s):
Machine vision, Pattern recognition, Least-square fitting, Algebraic distance.
Related
Ontology
Subjects/Areas/Topics:
Image Processing
;
Informatics in Control, Automation and Robotics
;
Mobile Robots and Autonomous Systems
;
Robotics and Automation
;
Vision, Recognition and Reconstruction
Abstract:
Real-Time tracking of elliptical objects, e.g. a ball, is a well studied field. However, the question between a monocular and binocular approach for 3D objects localization is still an open issue. In this work we implemented a real-time algorithm for 3D ball localization and tracking from 2D image ellipse fitting with calibrated cameras. We will exploit both approaches, together with their own characteristics. Our algorithm features the following key features: (1) a real-time video segmentation by means of a Gaussian mixture descriptor; (2) a closed-form ellipse fitting algorithm; and (3) a novel 3D reconstruction algorithm for spheres from the 2D ellipse parameters. We test the algorithm’s performance in several conditions, by performing experiments in virtual scenarios with ground truth. Finally, we show the results of monocular and binocular reconstructions and evaluate the influence of having prior knowledge of the ball’s dimension and the sensitivity of binocular reconstruction
to mechanical calibration errors.
(More)