Visual And Inertial Based Pedestrian Indoor And Outdoor Positioning Method

High precision positioning has become a common demand in many fields.Visual positioning and inertial positioning are both low-cost positioning methods who are independent from any infrastructures.The precision of vision method is high but the calculation time is pretty long.The inertia method is easy to realize but the precision is not high.Therefore,this thesis combines the video and inertial data of smartphone to achieve a high-precision and continuous pedestrian positioning independent of any infrastructure in indoor and outdoor scenes respectively.In the indoor scene,this thesis proposes a multi constraint image matching based localization method through combining indoor vision and inertia data who can realize the continuous and accurate indoor localization.This method uses smartphone to collect multi-sensor data,including video frames and inertial data.However,the computing time of continuous image retrieval and matching is a bottleneck of visual positioning.Therefore,in order to improve the computing efficiency of visual positioning method,a spatial model for optimizing the spatial structure of geographical marker image in database is proposed based on the spatial similarity calculation of multi constraint image matching.A series of image retrieval strategies based on spatial constraints are designed to further reduce the computing time.Based on the above spatial model and retrieval strategy under spatial constraints,a visual location algorithm is proposed.On the basis of image matching,a SFM method is used to calculate the direction change between video frames to correct the heading angle error in the inertial data.The calculation accuracy of PDR can be improved and realize high-density,high-precision and continuous positioning.At the same time,the results of inertial calculation assist visual positioning by optimizing image retrieval order to further improve the efficiency of image matching.The experimental results show that the average computing time of visual positioning is about 0.59 seconds,and the error of the method combined two algorithms is in the sub-meter level.So the indoor positioning with high accuracy,realtime and continuous is realized.In the outdoor scene,using the features of the ground road as a visual landmark,an outdoor blind location method based on the fusion of vision and inertia based on the extended Kalman filter is proposed.This method also only uses smartphone to collect multi-sensor data,including video frames,inertial data and continuous GPS coordinates.Based on the extended Kalman filter,GPS positioning results and PDR calculation results are combined to get a more stable and continuous pedestrian position.Based on collected road network and dataset,training the Yolo V3 model to detect the features of the ground road in the captured video.The visual landmark locations are applyed to modify the fusion positioning results.The experimental results show that the positioning accuracy of the fusion positioning algorithm is about 5 meters,which is obviously better than GPS positioning error and can better improve the accuracy of outdoor positioning.