Deep-Hart: An Inference Deep Learning Approach of Hard Hat Detection for Work Safety and Surveillance

摘要

The most common cause of injuries in the construction site was caused by falls, slips, and trips. As a response to the Occupational Safety and Health Administration (OSHA), this agency conducted training such as fall prevention. Despite these initiatives, there are still incidents and accidents that happened on the site. According to the study conducted by previous researchers, those fatalities can be reduced by wearing a hard hat. That is why OSHA requires all construction sites to strictly implemented the wearing of hard-hat within the vicinity of the construction site. This study developed a hard hat detection system to determine if the worker is wearing a hard-hat properly. Image processing was used in this study. The proponents used the public datasets with hard hat-wearing images to evaluate the performance by using the mean average precision (mAp) where the proponents obtained an average accuracy of 79.246. The proponents of the detection system of hardhats concluded that regardless of their size, color, types, and angles with an average Training and Validation accuracy of 97.29 and 92.55, average evaluation accuracy of 79.24% with the highest model accuracy of 86.89%, and testing accuracy of 86.67%. The system works properly.