Research And Implementation On Motion Recognition Based Indoor Pedestrian Positioning

With the rapid development of Location-Based Service and the rising of mobile internet,people's demands for indoor localization and navigation increases day by day.Since GNSS(Global Navigation Satellite System)cannot provide the positioning information in indoor environments because of the degrading or denying of the satellite signals,indoor positioning has been a research hotspot recently.And with the popularity of smartphones which are embedded with micro electro mechanical sensors,the indoor positioning method based on micro inertial sensors receives much concern.This paper focuses on how to implement a precise indoor positioning system using smartphone sensor.Because of the inevitable precision drift existed in micro inertial sensors,the positioning error will accumulate quickly along with time.This paper proposes the method of using human motion recognition to assist the pedestrian dead reckoning.With the recognized human motions,different step length parameters are applied.Thus,the indoor positioning precision can be promoted.And by recognizing the motion patterns in vertical direction,pedestrian dead reckoning can be extended to the three-dimensional space from the conventional two-dimensional plain.This paper mainly focuses on the following three research contents:1)The continuous sequence based human motion recognitionThe motion of a pedestrian can be seen as a continuous procedure which is consist of multiple basic motion patterns.There are relationships between two adjacent base motion patterns.This paper employs the Hidden Markov Model to describe the pedestrian motion procedure.It uses the motion state transition probability and observed variable probability to solve the motion recognition problem.And it can promote the rate of motion pattern recognition.2)The incremental learning based human motion recognitionIn the beginning of motion recognition,there are limited instances for training and the motion recognition rate may not be high.When the new training date comes continuously,the human motion classifier needs to be updated.This paper uses the incremental learning method to complete this work based on the original classifier and keep it performing well.3)The human motion recognition based pedestrian dead reckoningThe conventional pedestrian dead reckoning can only work in the two dimensional plain.And the step length model parameters are kept constant.Using human motion recognition,an adaptive step length model can be applied.The indoor positioning precision can be promoted.And by recognizing the motion state in vertical direction,the three dimensional pedestrian indoor positioning can be implemented.All in all,this paper focuses on the researches of the continuous human motion recognition,the incremental learning based human motion recognition and the motion recognition assisted pedestrian dead reckoning.And a completed indoor pedestrian positioning system based on smartphones sensors is implemented.