FAirWrite-Movement Reconstruction and Recognition Using a Low-cost IMU

We present a system for detecting air-writing motions conducted with a finger worn IMU. The vision is that a person could casually leave virtual notes in the environment by air-writing short notes when passing by a location. The challenge that we address is that such casual finger writing, in particular without a visual reference feedback (the user doing the air writing does not see the result), tends to produce extremely noisy trajectories which are inherently challenging to decode. The main contribution of this work is the development and systematic evaluation of a system for reconstructing such noisy trajectories and mapping them onto digits and characters. We describe the FAirWrite system's architecture, a Graphical User Interface (GUI) to interact with it, and the recognition and classification process to achieve the state-of-the-art results.