Current capacity and what we are developing


Deep Learning for Object Identification and Recognition in Challenging Environments

The aim of the project is to develop an autonomous system capable of real time object detection and recognition. The system should also be able to generate volumetric reconstructions about the objects it recognises. This capability is required to make the process of nuclear decommissioning more efficient, since the facilities that the proposed system will be used in are typically difficult to access and are currently inspected manually or by teleoperated machines. Automating this process reduces the hazard to human life while making the process a lot faster.

The project will focus on developing the volumetric reconstruction capabilities of the system using low cost depth sensing devices, while utilising state-of-the-art techniques for object detection. Current reconstruction and inference techniques are primarily based on using information from 2D cameras which does not provide actual geometric information about the detected objects. Some approaches use a multi-view paradigm which, though useful, is not always possible, especially for inaccessible environments.