Modified parallel differential evolution algorithm with local spectral feature to solve data registration problems

A novel parallel 3-D point data registration algorithm based on differential evolution with local geometric feature on GPU architecture was proposed. Firstly, feature points were extracted by local spectral descriptor from original data sets, and then correspondence between them was computed, at last, transformation matrix was calculated witch differential evolution strategy. In order to improve the matching speed of large amount of data, this registration algorithm was implemented in parallel pattern on GPU. The speed comparison and noise simulation experiments show that this algorithm is efficient and robust against noise.