A dataset to test AI on forensic scenes

Abstract

The aim of this article is to describe a dataset derived from crime scenes or fictitious accidents that can be used for artificial intelligence processing. The steps involved in setting up these forensic scenes, the data acquisition chain, classification and formatting will be described, as well as the first tests carried out in artificial intelligence. The first data formatting operations will be carried out so that they can be integrated into machine learning and deep learning solutions; in particular the deep learning solution SuperPointTransformer (Robert et al., 2023). The difficulties linked to this type of work and the benefits of using artificial intelligence for forensic scenes will also be presented.