TinyPPE: Ensuring Workplace Safety Helmet Compliance with Tiny Machine Learning

• Derek N. A. Alves Edge Innovation Center / Vertex Institute of Innovation and Technology / UFAL
• Erick De A. Barboza UFAL
• Tiago F. Vieira Edge Innovation Center / Vertex Institute of Innovation and Technology / UFAL

Resumo

Ensuring safety in workplaces such as industry and civil construction requires the use of personal protective equipment (PPE). Enhancing compliance with PPE regulations or employing educational measures through a deployable device, which can also restrict access to hazardous areas without proper safety equipment, is crucial to mitigating workplace accidents and reducing the risk of fatalities or disability. This paper proposes the development of a device equipped with embedded machine learning models that use a camera to monitor workers and enforce the usage of PPE. These models are deployed on an Arduino Nano 33 BLE Sense, integrated with an OV7675 camera shield, to capture images and employ our trained computer vision model to classify the usage of safety helmets. The compact size of the trained and quantized model, weighing only 59 KB, taking about 595 milliseconds total time, 92.5% accuracy, demonstrates its efficiency. This innovation not only promotes the adhesion to safety protocols, but also presents significant benefits in scientific and economic contexts.

Palavras-chave: Tiny Machine Learning (TinyML), Personal Protective Equipment (PPE), Computer Vision, Embedded Systems, Safety Helmet, CNN