Continuous Recognition of 3D Space Handwriting Using Deep Learning

In this paper, we attempt to present novel input methods that help enable byzantine free of hands interface through recognition of 3D handwriting. The motion is detected wirelessly by the use of the inertial measurement unit (IMU) of the Arduino 101 board. Two different approaches are discussed. One approach is to use the pattern matching engine (PME) of the Intel® Curie™ module on Arduino 101 mounted on the back of the hand. The second approach uses the IMU input to a well-structured recurrent neural network. The spotting of handwriting segments is done by a support vector machine. The former approach, being indigent of memory, is not preferred over the latter. The deep learning approach can continuously recognize random sentences. The model was trained on 1000 freely definable vocabulary and was tested by only one person, achieving the lowest possible word error rate of 2%.