Real-time surface reconstruction through HRBF networks

A procedure for 3D surface reconstruction from range data in strict real-time is presented. The process is based on a connectionist model, the hierarchical radial basis function (HRBF) network, which has been proved to be effective in the reconstruction of smooth, space varying surfaces. An efficient data structure and locality assumptions allowed the derivation of a faster configuration algorithm, without degrading the approximation capabilities. Results show that computing time scales linearly with the number of neurons, and is comparable with the data acquisition time of most commercial scanners so that reconstruction can be performed in real time.