Adaptive Differential Evolution Applied to Point Matching 2D GIS Data

The impetus behind data analytics and integration is the need for greater insight and data visibility, but since a growing share of our data is multi-media, there is a parallel need for methods that can align that data. This paper explores registration and referencing techniques that are used to combine spatial datasets and are used here to georeference map images to 2D GIS models. The intent of this effort was not to discover novel or optimal solutions to this problem, but rather to gain and share insight into the design choices and tradeoffs between functional and nonfunctional requirements to such problems. Accordingly, this paper provides a cross section of the dominant means, their strengths and weaknesses, and trajectory of research in this domain. More detailed treatment is given to knowledge representation, objective function, and the computational frame-work used to solve registration and referencing problems. The implementation presented here uses Hough transforms, nearest neighbor correspondences with simplistic similarity measures, and a population based optimizer. The comparison among meta-heuristics has shown that Differential Evolution (DE) frameworks appear especially suited for this problem. In particular, the controlled randomization of DE parameters appears to display the best performance in terms of execution time and competitive performance in terms of function evaluations even with respect to more complex memetic implementations.