Automatic detection of helmet usage in the construction site with deep learning methods

The law in force in Turkey to protect the health and safety of employees is the Act No. 6331 on Occupational Health and Safety. The measures to be taken by the workplace within the scope of this law begin with collective protection measures. However, due to the diversity of work items and the difficulty of applicability on construction sites, workers often have to work individually in different regions and dispersed settlement. For this reason, the necessity of protecting employees with the use of personal protective equipment arises. Helmet, which is one of the personal protective equipment, must be used legally by the workers in order to protect the workers from possible disasters in dangerous work areas. Similarly, according to this law, the employer has to monitor whether the helmets given to the employees are worn and warn those who do not. However, the large and dispersed construction sites make it difficult to follow-up employees. Thanks to the developing technology, automation systems are suitable for such follow-ups. In this study, it was determined whether the employees wear helmets or not, based on deep learning. For this purpose, two different solutions were followed, namely the deep learning-based image classification method and the image segmentation method. In the study, helmet detection was diversified with different analyzes by taking the whole body integrity of the employees and only their heads as a reference. When the outputs obtained were evaluated by comparing, it was seen that the analyzes focusing only on the heads of the employees were more successful. The image segmentation method was found to be more efficient among the two methods followed.