Evolving principal component clustering for 2-D LIDAR data

Abstract

This paper is accompanying the proposed implementation of the updated Evolving Principle Component Clustering (EPCC) algorithm for segmenting LRF (laser range finder) measurements into linear prototypes. The paper describes the target application for the algorithm, the algorithm itself and its implementation in C++ using Qt framework. The implementation is provided for both the proposed EPCC algorithm as well as for the popular split-and-merge (SAM) line segmenting algorithm and comparison is given in terms of computational complexity and results quality. The evolving nature of the proposed algorithm is most expressed in clustering approach itself and an on-line adaptation of cluster membership thresholds based on data observed in the past. The results conclusively show improvement over SAM in both the processing load and its stability in terms of low variations in how long the algorithm take to cluster various data sets.