Research And Implementation Of Pedestrian Dead Reckoning Algorithm Based On Inertial Sensors

The pedestrian dead reckoning is often used to estimate the position of user in the indoor environment.However,the deviation of the sensor measurement and the accumulation error in the traditional PDR algorithm have great influence on the final positioning result.Secondly,it is difficult to establish the exact relationship between the sensor data and the mobile placement.So it's not easy to classify them according to the data characteristics.Besides,the pedestrian walking trajectory may be a large deviation,crossing through the wall which should not exist.In order to solve the above problem,this thesis proposes an improved pedestrian dead reckoning algorithm.First of all,we propose a step count method through zerocrossing method with rising and descending state.Then,we use five variables: gender,height,step frequency,peak acceleration,valley acceleration,proposes a new step length calculation algorithm.The model is obtained by training the male and female data samples by multi-variable linear regression.We use particle filter to fix the step length of the intermittent jumping.Then,according to the linear correlation between the direction angle of the magnetometer and the relative steering angle of the gyroscope,the direction estimation model based on the angle variation of these two angles is obtained by linear fitting,which compensate the deviation through single magnetometer or gyroscope.In addition,for the five different device placement,we collect the acceleration signal of each way.The wavelet transform is used to extract the motion features,singular value decomposition is carried out by dimensionality reduction,and SVM is used to classify the training sets.Above methods help us determine the classification.And the step count and step length is recorded based on the characteristics of the user's insensitivity to the navigation trajectory in specific manner.Finally,this thesis proposes an indoor map matching algorithm,which first abstracts and models the interior space,and then restricts the user's trajectory through the points of interest,crossing-wall detection and direction correction.The result shows the trajectory is consistent with the real route.The experiment shows the results of different phase.In step count phase,the average error of the zero crossing and peak detection are 0.8% and 11.6%.The proposed step length error is 3.5%,which has a significant increase compared with the height-frequency model error 8.63% and the peak-valley root model error 10.84%.The 90% of sample direction error is within 20°,which has obvious advantages compared to the traditional single magnetometer and gyroscope.The accuracy of SVM is 95.62%,while the accuracy of Bayesian classification is only 82.31%.In the last map matching experiment,the trajectory is overcome by large drift and crossing the wall.The average error of the improved PDR algorithm is 1.48 m,and the trajectory can reflect the walking route better.In this thesis,the improved PDR algorithm with map matching,which greatly improve the positioning accuracy and have high practicability in indoor positioning.