3D Computer Vision

Three-dimensional object modeling and recognition on range data and 3D point clouds is becoming more important nowadays. We are exploring and studying new solutions and algorithms based on 3D point clouds, ranging from industrial applications to human computer interaction.

Current research activities on 3D Computer Vision

- 3D object reconstruction from multiple point clouds

- 3D object detection on large point clouds

- Hough Transform for sphere detection

Hough Transform for sphere detection

Hough Transform is a powerful pattern recognition method for parametric shape detection, and is widely used for 2D images. In this research activity we studied different probabilistic/randomized Hough Transform algorithms adapted for sphere detection on 3D point clouds, providing an appropriate formalism to describe their models. We also made a systematic performance analysis of these variants both on synthetic and real clouds. Synthetic data are available for further tests. Lastly we proposed a novel combined method which exploits the advantages of two different variants and performs better in terms of time and accuracy with respect to the variants taken individually, especially with clouds heavily affected by noise and outlier data.

Publications

1	Camurri, Marco; Vezzani, Roberto; Cucchiara, Rita "3D Hough transform for sphere recognition on point clouds" MACHINE VISION AND APPLICATIONS, vol. 25, pp. 1877 -1891 , 2014 | DOI: 10.1007/s00138-014-0640-3 Journal