First milestone (may 2003)

At first MAVR at DSI has focused on building the software engine able to filter the noise in the range data acquired by robot sensors and to transform them into 3D models. This problem has been tackled by optimizing for our problem, a particular neural network model developed by our group, called Hierarchical Radial Basis Function Network. Optimization is based on defining a set of operations, which can work locally on the data. The introduction of an efficient range partitioning schema has allowed to produce a very fast algorithm. Moreover, the developed technique allows to produce a model at different scales, feature which can be extremely useful in wire-less low bandwidth comunication.
For the first milestone, the software which implements this model has also been developed and it is available at the software WEB page of MAVR: homes.dsi.unimi.it/˜borghese/Research/Software/Software.html.

References

  • Borghese N.A., Maggioni M. and Ferrari S., Multi-Scale Approximation with Hierarchical Radial Basis Functions Networks, IEEE Trans. Neural Networks, In press.

  • N.A. Borghese, S. Ferrari and V. Piuri (2002) Real-time Surface Reconstruction through HRBF Networks. Proc. IEEE International Workshop on Haptic Audio Visual Environments and their Applications, Invited Paper. Proc. HAVE 2002, pp.19-25.
  • Ferrari S., Borghese N.A. and Piuri V. (2001), Multi-resolution Models for Data Processing: an Experimental Sensitivity Analysis, IEEE Trans. on Instrumentation & Measurement.  Vol. 50(4), pp. 995-1002.
  • Borghese N.A. and Ferrari S. (2000), A Portable Modular System for Automatic Acquisition of 3D Objects, IEEE Trans. Instrumentation & Mesurement , Vol. 49(5), pp. 1128-1136.