Project Metior Telum (“measure the weapon“) is a large scale AiLECS initiative to build a highly accurate automatic firearm detection system. A key part of this project is the construction of a hybrid machine learning dataset, which augments real image data with rendered firearm imagery obtained by 3D scanning and photogrammetry of real-world AFP weaponry holdings.
The AiLECS lab has engaged a professional vendor with many years experience in 3D scanning and photogrammetry for data capture work in this project. The dataset construction is being undertaken through cloud-based rendering of the captured firearms under a large variety of lighting conditions, geometrical orientations, occlusions and compositing onto various backgrounds. Regard is also being given on an application basis to ensure that synthetic dataset generation appropriately corresponds to real-world operational image feature distributions. A variety of deep learning models are being constructed for tuning and testing on this bespoke dataset.
Machine learning pipelines and models are being developed for this project, with the firearm scanning itself expected to commence in Canberra in April 2022. Technical and briefing notes for Metior Telum will be available shortly.