A floor-plan-free map matching method for foot-mounted IMU based pedestrian localization using visual SLAM

Abstract Indoor pedestrian localization plays a crucial role in ensuring the safety of personnel, particularly those working in hazardous environments, such as emergency responders, firefighters, and inspectors in dangerous areas. Accurate, real-time positioning can be life-saving in critical situations. However, due to the use of low-cost micro-electromechanical system (MEMS) inertial measurement units (IMUs), the positioning accuracy typically degrades over time. To address this issue, leveraging existing indoor floor plans and applying map-matching techniques has proven to be an effective way to improve accuracy. However, indoor floor plans are not always available, and updating existing plans can be challenging. In this paper, we innovatively introduce the simultaneous localization and mapping (SLAM) technique, originally developed for robotics, into the field of pedestrian localization. Specifically, we employ visual SLAM to construct a 3D sparse feature map, from which we generate a real-time 2D grid map. This grid map probabilistically represents the traversable areas of the indoor environment, effectively serving as a model of the indoor map. Once this 2D grid map is available, it can be used to enhance pedestrian localization accuracy through map-matching techniques. To validate the effectiveness of the constructed 2D grid map, we conducted pedestrian localization experiments in multiple indoor environments using a conditional random field-based map-matching algorithm developed in our previous work. The experimental results demonstrate that the proposed approach significantly improves long-term localization accuracy.