Fall Detection and Safety Accessory Monitoring System using Deep Learning

In industrial atmosphere, fall detection and safety accessory monitoring systems are crucial for automating tasks and enhancing worker safety by ensuring proper Personal Protective Equipment (PPE) use. Traditional methods involve streaming video data to centralized cloud-based systems for PPE monitoring, but this relies heavily on bandwidth and is vulnerable to network outages. Utilizing deep learning techniques like the YOLO algorithm with 95% accuracy, the system detects falls and monitors PPE adherence. Four YOLO algorithm versions are compared based on performance metrics. The model, trained on Google Co-laboratory, is deployed on a Raspberry Pi for onsite implementation, providing reliable notifications and fostering a secure work environment.

I. Introduction

In industrial environments, ensuring the safety of workers is paramount. Accidents, especially falls, can have severe consequences, underscoring the importance of proactive safety measures. By using deep learning techniques through the YOLO (You Only Look Once) algorithm, this system is capable of accurately detecting falls and monitoring workers' compliance with safety measures such as wearing helmets and jackets. Additionally, the system's deployment on a portable platform like the Raspberry Pi facilitates onsite implementation, ensuring seamless integration into industrial environments. The dataset comprises of 5,000 images. These images undergoes several preprocessing and augmentation steps and are trained using Roboflow and the model is trained on Google Colab and deployed on a Raspberry Pi for seamless portability and onsite implementation. The evaluation metrics are compared for four different versions of YOLO algorithm.