Research On Indoor Localization Method Based On Image Retrieval And Dead Reckoning

Obtaining accurate and robust location information in complex indoor environments has always been a hot topic in the field of positioning.With the development of micro-electromechanical technology,dead reckoning,which uses inertial sensors for indoor localization and can gain high accuracy,robustness and convenience of no need for deploying additional equipment,has become a hot topic in the field of indoor localization.Pedestrian dead reckoning is an autonomous and relative localization algorithm,which uses outputs of accelerometers,gyroscopes and magnetometers to make estimations of a pedestrian.It determines the relative position change of a pedestrian mainly through step detection,stride length estimation and heading calculation.However,due to the cumulative error of inertial sensors,the localization error of pedestrian dead reckoning will gradually increase when working continuously for long periods.In addition,traditional pedestrian dead reckoning can only update positions in the same plane and can not deal with the problem of positioning in threedimensional space,which have obvious limitations in practical applications.As an image retrieval technology,visual place recognition is also an autonomous positioning method and can be used to solve problems in pedestrian dead reckoning.We study pedestrian dead reckoning algorithm and train a special visual place recognition neural network on an indoor scenes dataset.A vision-aided pedestrian dead reckoning algorithm is proposed,which is compared with traditional pedestrian dead reckoning algorithm in an actual indoor scenes experiment.The localization error of the fusion algorithm proposed in this thesis is about 0.68% of the walking distance,and the localization accuracy is improved by 57.7% compared with pedestrian dead reckoning algorithm.Moreover,the fusion algorithm can successfully deal with the location change of pedestrians such as walking up and down stairs.