A system for collecting labelled data and evaluating AI algorithms for in-ear detection in hearing protection

Abstract

This thesis investigates the integration of artificial intelligence (AI) algorithms to precisely determine the status of an earplug within hearing protection devices, discerning whether it is correctly inserted in the ear or not. Motivated by the global prevalence of noice induced hearing loss and the need for effective protection, the study holds the potential to contribute towards increased safety in noisy work environments. Working with Minuendo, an innovative hearing-protection solutions company, it is addressing the issue by adopting AI in practical context.

Even though AI is becoming increasingly popular, its introduction into industrial settings presents multiple challenges. This thesis addresses these challenges by designing and implementing a system on cloud infrastructure, using modern practices that streamline the setup of the necessary architecture. This allows for straightforward experimentation with machine learning algorithms, supporting iterative development and ensuring reproducibility of results. The study merges theoretical insights with real-world applications, facilitating the integration of cutting-edge technology into industrial environments.

A significant aspect of this thesis is the exploration of various machine learning algorithms for detecting the in-ear placement of earplugs. Through initial analysis, strategic data gathering, and the pilot implementation of diverse machine learning algorithms, this research provides initial insights into their effectiveness and practicality.