Abstract
Research in field robotics often utilises scanning range sensors to aid autonomous navigation. The article addresses reliable feature extraction from continuously scanning range sensors operating outdoors. Contrary to other detection methods, an algorithm is presented which detects features on-line, as soon as the range to that feature has been sensed. A model is derived which makes predictions of range, before each new range sample is recorded. These are used to produce validation regions within which each new sample should lie, provided it belongs to the surface with the same smoothness characteristics as its range predecessors. The detection process model adapts its validation region according to the spatial gradient of the surface being sensed, and is implemented in extended Kalman filter (EKF) recursive form. Results are demonstrated with laser detection and ranging (ladar) sensor data recorded outdoors.