A Vision-Aided PDR Localization System

Pedestrian Dead Reckoning (PDR) is widely applied in indoor localization. Traditional PDR algorithms suffer from drifting errors. Filtering-based multisensor fusion techniques are proposed to alleviate this shortage. However, most of them require infrastructure built and pre-calibration for each individual environment. We hereby propose a vision-aided PDR system under a PDR-visual constraint model. General graph optimization framework is introduced to solve the problem. Its performance is validated on a public dataset.