Research On Pedestrian Indoor Inertial Navigation Algorithm Based On Human Motion Posture Recognition

Compared with Bluetooth,WIFI and other indoor positioning technologies,inertial navigation technology has attracted wide attention in indoor positioning scenes in recent years because it does not need to deploy basic equipment in advance,has fast positioning speed and is less affected by environmental factors.Most of the existing indoor positioning systems based on inertial sensors are only suitable for 2D and 3D dead reckoning positioning in normal walking state of human body,and can’t meet various pedestrian positioning with complex motion postures in real indoor life scenes.Therefore,this thesis deeply studies the positioning defects of the existing inertial navigation technology in the scene of multiple human movements,and builds an indoor 3D positioning system which is suitable for more human movements,higher positioning accuracy and stronger robustness.Firstly,the acceleration of common human motion postures is collected and analyzed.Aiming at the difference of acceleration waveforms of different postures,a step detection and feature extraction method based on buffered sliding window is proposed.Secondly,based on the classic feature ranking algorithm,a feature ranking algorithm based on multi-information fusion is proposed,which has better evaluation effect for human motion features.Then,in order to realize the complex motion posture recognition of human,the multi-classification algorithms such as support vector machine,random forest,Adaboost,KNN,etc.are implemented and the classification accuracy is compared,and the best classifier with low feature dimension,high classification efficiency and more accurate classification is selected.Then,according to the defects of the existing methods of dead reckoning for human body in the application of multi-motion postures,some improvements were made in such aspects as step size estimation,height calculation and pseudo-step elimination.Through the dynamic adaptive step size model,we estimate the step size of multi-motion attitude,which reduces the positioning error caused by the step size estimation error.Aiming at the problem that the existing height estimation algorithms are affected by the environment such as temperature and air pressure,this thesis takes the height value calculated by the fusion of barometer and vertical acceleration Kalman as reference,and puts forward a height estimation algorithm with higher accuracy and stronger robustness by integrating the information of human motion and posture.Considering that the existing step detection method based on acceleration is easy to be misjudged due to the influence of in-situ pseudo step motion,which leads to positioning error.Based on the study of indoor magnetic field variation law,a pseudo-step elimination algorithm is proposed to eliminate in-situ movements such as turning,stepping and shaking.Finally,the indoor three-dimensional dead reckoning algorithm is tested for human positioning in two-dimensional plane and three-dimensional space,and the pseudostep elimination effect is tested.To sum up,this thesis proposes and implements a three-dimensional positioning algorithm suitable for the scene of multiple human movements and postures.Experimental results show that the recognition algorithm in this thesis can identify 7 kinds of motion postures with an accuracy of more than 99% with a collection of 8 witters,and the relative errors of 2D and 3D positioning of the positioning algorithm in indoor human body with multiple motion postures are less than 1%,which is more suitable for the real use environment.