Differential evolution algorithm based on CUDA and its application in unorganized points data registration

Automatic 3D unorganized points data registration technique, which maps data clouds measured from multiple viewpoints into a common coordinate space, is an key teclinical on reverse engineering. In order to improve the matching speed of large amount of data, a detecting method with parallel diffrential evolution algorithm based on CUDA architechure was proposed. Firstly, evolution strategy was proved, to avoid premature convergence and improve optimizing speed, the probabilities of crossover and mutation were adaptively adjusted by means of adaptive algorithm, Secondly, parallel-DE-matching algorithm model was designed according to feature of GPU. The comparison experiments show that this method is effective and efficient for aligning large number of three dimension clouds data.