Modeling of Structure Landmark for Indoor Pedestrian Localization

The decreasing of accumulative error is a key issue for various multi-sensor fusion-based indoor localization systems that employ pedestrian dead reckoning (PDR) to improve their localization performance. Current studies mainly use activity-based map matching (AMM) to prevent the accumulative error. However, it is vulnerable to mismatch problems, which are usually caused by the randomness of human activities. This paper proposes a structure landmark map matching-based indoor localization approach. Structure landmarks refer to special spatial structures (e.g., intersections, corridors, or corners), which are visually salient in a local environment. These landmarks are visually recognizable in indoor spaces because of their distinct shapes. This paper integrates visual and inertial information to recognize the structure landmarks by using a Bayesian classifier. An algorithm is also proposed to realize indoor localization without prior knowledge of the initial location or the turning angles of people. This approach decreases the accumulative localization error of PDR by matching the detected structure landmarks to the ground-truth values. The experimental results showed that the identification accuracy of the structure landmark was about 90% and the matching accuracy was 92%. The mean off-line localization error was about 1.2 m. Compared with the AMM-based method, this approach is robust to the random turning activities of people and can realize indoor localization with a faster convergence speed.