This paper presents an approach to extract human walking paths independently from the orientation of the paths in a global coordinate system. Previously, observing human walking, connectivity between the spaces (areas) has been obtained. In this paper, we regard human walking paths as a feature to represent patterns of activities. Observing and describing human activities can be considered as useful information for intelligent environments to enable the environments to provide suitable support to the users corresponding to their actual situations. In this paper, we present an approach to extract human walking paths independently from the orientation of the paths in a global coordinate system. More specifically, we propose a similarity measurement based on AMSS (Angular Metrics for Shape Similarity), then classify human walking paths using a hierarchical clustering method. Experimental results show that the proposed approach achieves rotation invariant extraction of human walking paths.