Object detection in a framework for automated nuclear waste classification

In this paper, we present a new framework for automatically triaging nuclear waste classification inside a nuclear cell for decommissioning. The process of decommissioning includes a large amount of human involvement for decision making, physical inspections and even lifting and relocating radioactive waste items. The current process accounts for risks like close human contact with radioactive material for extended periods of time, and errors based on operator knowledge rather than automated detection systems. The aims of this new framework are to reduce cost and speed up the sort and segregation process by providing a list of expected waste items, their location within the cell, and expected waste classification autonomously. We aim to reduce the reliance on the subjectivity of human decisions by capturing, formalizing and codifying their knowledge and experience and reducing the potential for errors arising from reliance on individuals. This paper focuses on the design and description of the framework and demonstration of the first step of the framework through a case study drawn from a mockup of a nuclear cell. We perform planar segmentation and cylinder detection on a point cloud dataset using RANSAC based methods to, firstly, distinguish indoor walls from objects for processing, and to detect objects and their estimated parameters.