Indoor Pedestrian Location Estimation with HAR Assistance

Human activity recognition (HAR) technology plays a crucial role in inferring indoor environmental information through the analysis of pedestrian activity data within indoor environments. However, traditional methods that utilize recognized environmental information as landmarks for indoor localization face limitations in map dependency and timeliness issues. To overcome these challenges, we propose a novel approach that leverages deep learning techniques to estimate both position and activity recognition in real time directly from raw inertial sensor data. In contrast to the traditional approach of relying on indoor maps for location correction, our method directly utilizes the results of HAR identification for real-time position estimation, eliminating the need for pre-existing maps and ensuring optimal performance of the positioning system. To support our research, we constructed an indoor activity dataset consisting of six common activities: standing, walking, running, escalator up, escalator down, upstairs, downstairs. The experimental results indicate that the proposed method not only reduces errors but also ensures stable indoor positioning.