Protective glasses detection in occupational safety with deep learning

With developments in Deep Learning studies, more accessible use, and the development of computer vision technologies, real-time object detection systems have become widespread. Eye detection from images from object detection studies has become a significant operation for security, banking, courthouse, medical fields, driverless vehicle systems, occupational safety, and health. In this thesis, a study was carried out that detects eyes in real-time from images and detects eye protection and protective glasses in terms of occupational safety and health in enterprises. In this study, two different detection models were developed. In the first model, the eye was detected from the facial parts. In the second model, a model was developed to control the use of protective glasses by the employees in the workplace. This model can distinguish regular glasses and protective glasses from images. In the study, models were created by training with the graphics processing unit (GPU) configured on the computer with the datasets obtained specifically for working over the internet. The models were tested with different images, the performances of the model at different weights were compared, and the results obtained with the test were analyzed. The results obtained showed that the hybrid use of CNN networks and the YOLO algorithm gave successful results in the detection of limbs and goggles.