Knowledge discovery in choreographic data using Relative Motion matrices and Dynamic Time Warping Seyed Hossein Chavoshi a, * , Bernard De Baets b , Tijs Neutens a , Hyowon Ban c , Ola Ahlqvist d , Guy De Tré e , Nico Van de Weghe a a Department of Geography, Ghent University, Krijgslaan 281 (S8), Ghent, 9000, Belgium b KERMIT, Department of Mathematical Modelling, Statistics and Bioinformatics, Ghent University, Belgium c Department of Geography, California State University, Long Beach, USA d Department of Geography, Ohio State University, USA e Department of Telecommunications and Information Processing, Ghent University, Belgium Keywords: Motion pattern Moving point object REMO DTW Similarity analysis abstract Recent advances in location-aware technologies have led to the exploitation of geospatial methods to uncover valuable information from large movement data sets. The focus of this study is on choreographic data. In particular, the goal of the study is to visualize and analyze the motion patterns of samba dancers during their performance by means of two complementary methods. The first method performs map algebra with RElative MOtion (REMO) matrices to study the evolution of motion attributes, such as speed, motion azimuth, and vertical angle over time. The second method applies Dynamic Time Warping (DTW) to time series of motion attributes. The results demonstrate that both methods are useful in numerically comparing the performance of samba dancers and visually exploring the motion patterns of different body parts. Ó 2013 Elsevier Ltd. All rights reserved. Introduction Motion is the change in position of objects, such as vehicles, animals, hurricanes or oil spills, with respect to time. Recent technological advances in location-aware technologies, including Bluetooth sensors, RFID, 1 and GPS, 2 have made it possible to track changes in the positions of such objects and gather enormous amounts of movement data, often as a series of discrete observa- tions of moving objects represented as tuples of x, y, z coordinates and time t. This enormous amount of positional data has led re- searchers from different disciplines to develop new ways to explore movement data. Particular attention has been directed toward analyzing human movement. For example, Sigal, Balan, and Black (2010) developed a hardware system to capture synchronized video and ground-truth 3D motion. Chaudhry, Ravichandran, Hager, and Vidal (2009) presented an activity recognition method that classifies the human activities in video sequences. Nagashima and Katsura (2012) demonstrated a method for analyzing the principal components of human motion in time series and esti- mating the functional mode of the human motion. Yuan and Raubal (2012) developed a technique for analyzing dynamic mobility patterns of mobile phone data sets using Dynamic Time Warping (DTW). Observations of human motion may be related to individual or collective motion behavior (Andrienko, Andrienko, Kopanakis, Ligtenberg, & Wrobel, 2008), and human motion has been stud- ied in various research domains, such as social sciences, geo- marketing, transportation, political sciences, biomedical analysis, and sports and recreation. Due to the large amount of movement data that is becoming available through mobile sensors, data mining methods are necessary to uncover useful, hidden information that can be used for different purposes. In this study, we aim at understanding the motion patterns of moving point objects (MPOs) through similarity analysis. Because motion patterns are clearly visible in many rhythmic dances, we used some basic movements of samba as a case study to investigate two approaches to measure similarity in motion patterns, namely RElative MOtion (REMO) and Dynamic Time Warping (DTW). * Corresponding author. Tel.: þ32 (0) 487 248 335; fax: þ32 (0) 9 264 49 85. E-mail addresses: seyedhossein.chavoshi@ugent.be, hossein.chavoshi@gmail. com (S.H. Chavoshi), bernard.debaets@ugent.be (B. De Baets), tijs.neutens@ugent. be (T. Neutens), hban@csulb.edu (H. Ban), ahlqvist.1@osu.edu (O. Ahlqvist), guy. detre@ugent.be (G. De Tré), nico.vandeweghe@ugent.be (N. Van de Weghe). 1 Radio Frequency Identification. 2 Global Positioning System. Contents lists available at ScienceDirect Applied Geography journal homepage: www.elsevier.com/locate/apgeog 0143-6228/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.apgeog.2013.12.007 Applied Geography 47 (2014) 111e124