A new dataset obtained from a real CCTV installed in a university and the generation of synthetic images are presented, resulting in a weapon detection model able to be used in quasi real-time CCTV improving the state of the art on weapon detection in a two stages training.