Research On SmartPhone Based Pedestrian Dead Reckoning

With the rise of Location Based Service(LBS),more and more attrention has drawn to indoor localization and pedestrian tracking.Beacon free method like pedestrian dead reckoning gradually become the hotspot in this field.On the other hand,as the increasingly popular smartphone becomes the primary carrier device for personal location-based applications,the fact that smartphones are assembled with adequate sensor makes it possible to locate users in the room.By leveraging the MARG(Magnetic,Angular Rate,and Gravity)sensors assembled in smartphones,this paper realizes the design of a robust pedestrian dead reckoning in indoor environment.The main difficulty of this technique is how to achieve accurate pace detection and heading estimation under a complex situation where the location mobile phones is of different ways(which in this paper is summarized as three modes,namely,navigation mode,swing arm mode,pants pocket mode)and the pacing speed is changable.To address the difficulty mentioned above,we present solutions as following:Firstly,under the circumstance that the traditional window peak detection works badly to fast speed motion(eg,fast walking or running),this paper designs an adaptive window peak detection algorithm based on the relationship between the acceleration peak and the pace frequency at different walking rates to achieve robust step detection.Secondly,due to observation of wrong counting of steps caused by the "pseudo" false walking action which conforms to the peak feature of the step acceleration peak,we uses Dynamic Time Warping(DTW)to analyse the correlation between potential step signal divided by the forementioned adaptive window peak detection algorithm and a confirmed step signal to distinguish between walking and "pseudo" walking action.Then,the kNN step length prediction is used to modify the poor fitting of linear modle of step length in case where the step length data becomes large and variable.Finally,this paper uses the attitude quaternion expression and the Kalman filter to solve the attitude of the mobile phone as well as reducing the erorr of attitude calculation through sensor fusion.By doing that we can get the estimation of yaw angle in the navigation smartphone mode.Then,by taking use of the rule of orthogonality between the swing axis and heading horizontal and vertical in the swing arm and pocket model,and the fact that the direction of average horizontal speed between two steps is near to walking direction,we can estimate the heading under swing and pocket mode,achieving the heading estimation of non-constrained smartphone.Experimental results show that compared with the traditional step detection scheme,the accuracy and robustness of the proposed scheme are better,achieving an average detection accurancy of over 95%.the improved step length estimation model reduces the error of linear step length estimation method.The error of heading under various position of the mobile phone is within the acceptable range(90% overall error less than 20 degree);In the practical application,the results of overall pedestrian dead reckoning system are pretty good.