Air Writing Recognition System through Deep Learning

A person can engage in air writing by moving their body while they write letters or words on a blanksurface. An example of gesture recognition is "air-writing recognition," which involves turning handgestures into invisible handwriting. You don't need to commit certain gesture patterns to memory in order topractice air-writing, unlike with general motions. It focuses instead on the language and subject matter thatthe user is interested in. With the advent of smart bands, the need for an extra device with sensor(s) is nolonger an issue with conventional air-writing. Consequently, algorithms for air-writing recognition arebecoming increasingly flexible on a daily basis. However, signal length variability is one of the keychallenges in developing an air-writing recognition model. The uneven signal length is made clearthroughout this writing and data-recording process. To make sure the signals were always long, researchersused techniques like padding and truncating, but these approaches led to massive data loss. To ensureminimal data loss for time-series signals, a statistical technique called interpolation can be employed. Afterextensively investigating different interpolation methods on seven publicly available air-writing datasets, weutilized a 2D-CNN model to identify air-written characters in this study. All datasets showed that oursolution outperformed the state-of-the-art solutions in terms of user-dependent and user-independentprinciples.