Absolute localisation in confined spaces using deep geometric features

When operating in dark and confined spaces, the capacity of robots to localise an absolute reference frame is utmost importance. This letter presents localisation system, using deep 3D landmarks, for known spaces. The system estimates robot's relative visual inertial odometry. Local landmarks are extracted from view. Similar are, also, prior map which then registered with local provide via extended Kalman filter. To best knowledge, feature registration has not been used before localisation. proposed tested within a representative application area—i.e. structured, space—and results indicate greater accuracy lower processing time when compared mainstream approaches.