Algorithm Research On Personal Navigation System Based On Particle Filter

As a representation among numerous information technologies, the navigation technology is embedding into our daily life in all minutiae. How to adapt to complex environment and to merge multi-sensor information in order to achieve more accurate navigation is becoming the key problem in navigation field. Meanwhile, inertial navigation system avoids dependence on the signal source and flexible to use that make it becoming more and more important in personal navigation field. This paper proposes a kind of personal navigation system that making use of the inertial sensors and combining the machine learning support vector machine method and particle filter to eliminate the accumulated error.In this paper, by extracting steps heading angle changes and other information which are caculated by the system based on kalman filter, establish reckoning motion model, through the particle filter algorithm complete trajectory optimization. The trajectory optimization module based particle filter includes both of activities recognition and plane map information fusion:Firstly, the plane map information provides an important criterion for dead reckoning accuracy, using the planar map information, determine whether each particle for each step is correct, then eliminates the false particle through the secondary particle optimization that ensures the trajectories compliance with the objective facts; In addition, preprocessing inertial sensors and pressure sensor data preprocessing, including coordinate transformation, high-pass filter, the difference between the calculated pressure and so on, identified human activity by two levels of SVM which includes stationary, walking up and down stairs, down the elevator, the system try to extract a correct point from the activities involving geographic information that is identified, then apply it to the particle filter, In process of the particle projecting correct the positioning results.The experiment shows the plane map information fusion experimental evident system fixed error results through walls; Join the activity recognition correction module, when the result error of the navigation system based on kalman filter is large, its cumulative error could be controlled within2%.The study shows that combining flat map information to impact the update and transfer of weight of particle eliminates the error particles effectively; Meanwhile, support vector machine is used to recognize the human activities by the inertial sensor data that provides higher recognition accuracy. Then the correction system consists of the flat map involving geographic information and particle filter that improve the accuracy of navigation system; When the result error of the navigation system based on kalman filter is large, optimizing its cumulative error to make it to be controlled within2%, the algorithm has certain feasibility.