A Novel Pedestrian Dead Reckoning Method Based on Visual Feature Map Matching Constraints

Robust and precise pedestrian navigation in satellite denial environments is a challenging problem. Existing inertial based positioning solutions still suffer from large drifts in long-time operation. In order to achieve reliable inertial positioning without any extra infrastructure, this works proposes a novel pedestrian dead reckoning (PDR) method that constrains the inertial system drifts by exploiting visual feature map matching and graph optimization. In doing so, the cumulative error of PDR can be effectively corrected by optimizing inertial positioning results with vision poses from matching current visual observation against the given feature map. Our experimental results demonstrate that the proposed method can obtain decimeter-level pedestrian position results in several typical indoor and outdoor scenarios, and significantly improve the long-term positioning performance. The averaged positioning error is reduced by 81.3% compared with a pure PDR without our introduced map matching based visual positioning.