Integrating prior knowledge in time series alignment: Prior Optimized Time Warping

Abstract

In this paper, we propose Prior Optimized Time Warping (POTW) algorithm, which allows user to integrate prior knowledge by marking out pairs of matching sub-sequences from the sequences to be aligned. To relieve users of the task of guaranteeing the full accuracy of the marking, a certainty coefficient reflecting the certainty of the matching can also be specified for each marked pairs. POTW will then look for the best alignment based on the two sequences and the given matching pairs. POTW is an extension of existing align algorithm, and in the absence of prior knowledge, is able to independently find the best alignment of two sequences. We apply our algorithm to walk sequences from CMU motion capture database, as well as UJI pen characters dataset to demonstrate its ability to allow easy and effective integration of prior knowledge.