Geometric Compression of Orientation Signals for Fast Gesture Analysis Aswin Sivakumar † , Rushil Anirudh *† , and Pavan Turaga *† * School of Arts, Media Engineering † School of Electrical, Computer and Energy Engineering Arizona State University Tempe , AZ - 85287 Abstract This paper concerns itself with compression strategies for orientation signals, seen as signals evolving on the space of quaternions. The compression techniques extend classical signal approximation strategies used in data mining, by explicitly taking into account the quotient- space properties of the quaternion space. The approximation techniques are applied to the case of human gesture recognition from cellphone-based orientation sensors. Results indicate that the proposed approach results in high recognition accuracies, with low storage requirements, with the geometric computations providing added robustness than classical vector-space computations. 1 Introduction Wearable sensors today are used in a wide variety of applications like health monitor- ing, fitness tracking, and gaming on smart phones. Many of these applications involve various forms of activity or gesture analysis i.e., making inferences about the type of motion based on the recorded sensor data [1]. For reliable performance, such applica- tions require continuous sensing and recording sensor data, combined with computa- tionally intensive algorithms for inference, both of which can impose a heavy load on resources. This issue manifests, for instance, as reduced battery life on a cellphone while continuously executing activity analysis algorithms. The underlying inference computations can add further load when the data representation has a non-Euclidean interpretation. In the case of activity and gesture analysis, rotation or orientation information is usually represented in the form of quaternions or as rotation matrices, both of which have associated non-Euclidean geometric properties [2]. In such a case, one cannot use classical vector-valued signal approximation strategies, but one needs to take recourse to differential geometry, and extend classical signal approximation techniques with geometric computations. However, non-Euclidean computations are intensive and often iterative, thereby demanding increased resources. To alleviate this problem, in this paper we propose to use a symbolic approxi- mation method that is an extension of classical vector-quantization approaches used This research was supported by National Science Foundation grant number CCF-CIF 1320267. Corresponding author - ranirudh@asu.edu