Research Of Multi-mode Pedestrian Dead Reckoning System Using Smartphone

Due to the rapid development of communication technology and the widespread popularity of smart devices,the requirement of Location-Based Service(LBS)has greatly been developed.As the basis of LBS,positioning technology has also been rapidly promoted.The satellite positioning and navigation systems led by GPS have been able to provide continuous and accurate position information.However,in a relatively closed indoor environment,satellite signal is easily obscured by various obstacles such as walls during the propagation,which is one of the new challenges for the precise positioning service.Among the existing indoor positioning solutions,Pedestrian Dead Reckoning(PDR)technology based on inertial sensors has broad application prospects because it is free from the interference of the external environment and does not need to deploy additional facilities.Most of the existing smartphones have a variety of built-in sensors,and the indoor positioning technology based on these sensors has high scientific research value and practical significance.The existing researches on pedestrian indoor positioning are mostly based on fixed posture of pedestrian holding smartphones.However,in most realistic scenarios,the postures are random and convert momentarily during the pedestrian's movement.In order to effectively improve the practicality and precision of PDR algorithm,it is especially important to study the PDR algorithm under a variety of postures.Most smartphones run the Android operating system.In this work,a smartphone with Android system is used as the hardware device.Firstly,recognition method of five different smartphone modes are studied.The characteristics of data signals from accelerometer and gyroscope are extracted at the various smartphone modes.Then a multi-mode support vector machine classification model is trained to realize the accurate recognition of mode during the pedestrian's movement.Secondly,three parts of PDR algorithm,step count method,step length estimation method and orientation estimation method,are studied separately.For the study of step count method,the mode-related peak detection method is proposed,in which the acceleration component corresponding to the current mode is selected and then dynamic threshold determination condition is added under various modes.For the study of step length estimation,a frequencylength relation model is established by linear fitting method.For orientation estimation,the orientation is estimated by Kalman filter fusion gyroscope and magnetic sensor,which effectively exploits the advantages of both methods and improves the accuracy of orientation estimation.Finally,three parts of the PDR algorithm are integrated to realize the multi-mode indoor positioning system.By comparing the pedestrian's movement trajectory depicted by our positioning algorithm with the real trajectory,the performance of each part of the algorithm is effectively evaluated.The experimental results show that the accuracy of the mode recognition rate based on support vector machine is up to 96.82%;for the step count part,the step count accuracy is higher than 96% under fixed mode and higher than 95% under multi-modes;for the step length estimation process,the 20-step distance estimation error is less than 10% under different modes;at the orientation estimation phase,the error in orientation estimation of 90% of the sample is within 35 degrees.The multi-mode indoor positioning system proposed in this thesis can effectively improve the accuracy of indoor positioning and has high practical value.