A hybrid [ICP and GA] image registration algorithm for depth images

Abstract

Iterative Closest Point (ICP) is an algorithm used to find the rotation and translation to efficiently register two point sets. A major drawback of the ICP algorithm is that it demands the data point sets to be approximately registered before it can be applied. Genetic algorithms (GA) provide a global solution to this problem and have no such prerequisite, but their convergence speed is slow. In this paper, we have demonstrated the use of a hybrid of a binary genetic algorithm (GA) and the Comprehensive ICP (CICP) algorithm (an existing variant of the ICP algorithm) to register depth images of a human face. The application of the GA followed by the CICP algorithm has proven to be fast and efficient and has no precondition on initial registration. The hybrid algorithm was able to register twenty control points to an RMS error of 0.6148 in 2.0187 seconds on an average.).