21-2 Human Locomotion Segmentation by Constrained Hierarchical Clustering of GPS Data

Abstract

We present a system for segmenting continuously archived human locomotion data. A portable GPS receiver acquires and stores data regarding its bearer's location and motion. A constrained hierarchical clustering algorithm segments these data into two basic classes of locomotion. A visualization incorporating user knowledge allows effective summarization of the results. A quantitative evaluation demonstrates that the average accuracy of the clustering algorithm is approximately 94%, despite the presence of location-dependent noise.