Image-Based Localisation (IBL) provides a novel approach for estimating the geo-location of a mobile platform by matching a query image to a pre-collected catalogue of georeferenced scenes. This technique faces significant challenges, such as viewpoint variation and perceptual aliasing, particularly when matching scenes captured from orthogonal viewpoints. Previous research has demonstrated the use of image-based localisation to recover the geolocation of mobile platforms, primarily focusing on improving recall metrics [142, 54, 106, 124, 155, 38]. This thesis investigates the development, implementation, and improvement of an IBL approach, capable of operating in near-real time, for both aerial and ground level platforms. Combining neural networks, temporal filtering and efficient data structures, we develop a localisation technique that improves recall and robustness, which we validate on both mobile platforms. This thesis makes three contributions. First, it introduces the Cross-View Shared Transformer (CVST) network, which improves recall and retrieval accuracy for ground-level platforms operating with both panoramic and limited field of view (FoV) images. Second, it provides a novel localisation technique that enables an Unmanned Aerial Vehicle (UAV) to operate in rural environments under restricted GPS conditions. Third, it enhances the robustness of cross-view localisation in ground-level platforms by integrating the CVST with a particle filter and a 3D KD-tree data structure to achieve near-real time performance. Further improvement in robustness is achieved through a mathematical formulation of the particle filter likelihood that incorporates knowledge of the road network, thereby improving the localisation accuracy of the system.