Abstract
Vision based object detection is a key feature within planetary rover navigation which facilitates several functions such as hazard avoidance, localization and path planning. Most of the current research is based on stereoscopic vision or multiple cameras strategically placed along the rover chassis that perform one specific function. This works for large rovers with sufficient processing power, however such resources would not be very practical for small or micro-rovers. This thesis aims to extract terrain surface information from a single camera mounted on a micro-rover such as the Surrey Mobile Autonomy and Robotics Testbed (SMART) based on minimal computational resources. The terrain surface information can provide feature inputs to other on-board navigation functions such as the Planetary Monocular Simultaneous Localisation and Mapping (PM-SLAM) and constellation matching. The detected terrain surface can also be of scientific interest due of the geometrical characteristics produced from this research. This research aims to improve the processing speed of the Guidance Navigation and Control (GNC) system using low level 2D image processing techniques. The methods employed result in a faster "perception stage" of the GNC with lower processing power requirements, creating structural information, shape descriptors and cognitive segmentation/classification of the rover’s surrounding environment. Although the initial application of this research is for planetary rovers, the research outcome is envisaged to be relevant, and hence transferable, to other vehicle navigation problems used on land, air or under water.