We provide an object detection model that demonstrates the usability of deep learning in helmet detection. Two different datasets are utilized. The Open Image v7 dataset and a custom-made dataset. The custom-made dataset comprises three different scenarios that receive different outcomes. Each of the scenarios comprises three different cameras. The different cameras per scenario are all focused on the same place. Different resize parameter values are chosen based on the mean of the data and the maximum values. Results show that resizing to the mean values is the better option. Besides that, the results show that there is a significant difference between the cameras per scenario.