Non-metric constraints in the graph-based optimization for personal indoor localization

Graph-based optimization is a framework that makes it possible to obtain the most probable estimates of the system state based on the observed constraints. This approach is considered the state-of-the-art in the localization of autonomous agents and SLAM. Most commonly, metric (quantitative) measurements are added to the graph, as non-metric (qualitative) measurements are ignored due to the troubles in representing that information as valid graph constraints. We propose methods to include non-metric information in the agent localization process employing graph-based optimization. The proposed method is experimentally verified for the case of personal indoor localization using a mobile device. The non-metric information in this application is particularly important, due to the limitations of the sensors in a typical smartphone or tablet that do not allow the agent to gather enough quantitative information to constraint its pose. Hence, constraints that stem from the qualitative information can be used to improve the reliability and accuracy of localization. These constraints come from the processing the state of the mobile device, from the known environment 2-D map (blueprints), and from appearance-based place recognition. The proposed methods of representing non-metric measurements as graph constraints are verified in real-life experiments and are a guide how different types of non-metric constraints could be included in the graph-based optimization.