Level-Set Based Greedy Algorithm With Sequential Gaussian Process Regression for Implicit Surface Estimation

This paper proposes an efficient greedy algorithm to estimate the object shape within a limited number of sample data (i.e. the touch-down points). Specifically, we treat the object shape as an implicit surface which is defined as the zero level-set of an unknown function and apply Gaussian process to estimate the surface. The mutual information criteria is utilized to decide which point should be sampled in the next iteration. To expedite the estimation process, we implement sequential Gaussian process for computational efficiency and significantly reduce the computational cost by selecting search area based on the estimated level-set variance. We present some simulation results to compare the performance of the proposed algorithm with the random sampling algorithm and demonstrate the improvements in both speed and accuracy over the random sampling algorithm.