Miniature air vehicles (MAVs) are particularly well suited for short-distance, over-the-horizon, low-altitude surveillance and reconnaissance tasks. New camera and battery technologies have greatly increased a MAVs potential for these tasks. This thesis focuses on aerial surveillance of borders and roads, where a strap-down camera is used in-the-loop to track a border or road pathway. It is assumed that quality tracking requires that the pathway always remain in the footprint of the camera. The objective of this thesis is to explore roll-angle and altitude-above-ground-level constraints imposed on a bank-to-turn MAV due to the requirement to keep the pathway in the footprint of a downward-looking strap-down camera. This thesis derives the required altitude to maintain the pathway in the footprint of the camera and associated bank-angle constraints. Constraints are derived for both roads whose geometry is unknown a priori and roads with known geometry obtained from digital elevation map (DEM) data. MAV geometry and camera localization are used to derive these constraints. The thesis also discusses simple computer vision techniques for pathway following and a corresponding guidance law. The pixels of the captured color video are statistically classified into road and non-road components. Standard computer vision functions are used to eliminate classification noise and obtain a road heading direction. The effectiveness of the result is explored using a high fidelity simulator. Flight test results on small UAVs demonstrate the practicality of the road-following method.