Adaptive real-time dual-mode filter design for seamless pedestrian navigation

Abstract

Abstract-Seamless navigation requires that the mobile device is capable of offering a position solution both indoors and outdoors. Novel seamless navigation system design was implemented and tested to achieve this aim. The design consists of general navigation system framework blocks and of the necessary interface agreements between the blocks. This approach enables plug-and-play style design of modules.The implementation used four preselected key technologies. Microstrain 3DM-GX4-45 foot-mounted inertial measurement unit sensor data was fused together with the u-blox GNSS receiver positions outdoors. Context sensitive inference engine enabled the fusion of position updates indoors from the Decawave TREK1000 Ultra WideBand ranging kit and from the 6 Kontakt.io/Raspberry Pi anchor-based Bluetooth low energy fingeprinting system.Novel dual-mode filter design uses a particle filter and the pentagon buffer enhanced Kalman filter in the position solution derivation. Depending on the map and the walls in the environment and on the quality of position updates, the implemented control logic employs the most fit filter for the current context. Computational power is now focussed, when particle filter is needed. The novel pentagon buffer enhanced Kalman filter is 10 times faster, allowing power saving when situation is not too critical. Moreover, the buffer provides position updates by interacting with the map and helps to correct the position solution.The navigation system is seamless according to the tests conducted around and within the Nottingham Geospatial building. No user input is needed for smooth transition from outdoors to indoors and vice versa. The system achieves an accuracy of 2.35 m outdoors and 1.4 m indoors (95% of error). Inertial system availability was continuous, while GNSS was available outdoors and BLE and UWB indoors.

A tripartite filter design for seamless pedestrian navigation using recursive 2-means clustering and Tukey update

“…The particle_filter has 200 particles. This was found adequate to our need for accuracy (meter-level) in [3]. A particle state consists of x and y positions, direction of movement and weight values.…”

“…This is shown in yellow. done previously and described in [3]. The zero update (ZUPT) detection method checks the near real-time low pass filtered outputs of the total accelerometer magnitude using two thresholds (1,08g and 0.92g) and of the total gyroscope magnitude using one threshold (0.3rad/s).…”

“…These were presented previously in [3]. Line segment intersection method is used together with a method that checks whether a point is inside a polygon.…”

“…pentagon_buffer is a five-point error estimation buffer that is around the main solution of the Kalman filter and was also introduced in [3]. Fig.…”

“…The design can adapt solution computing resources and methods according to the inferred context. Four previously preselected sensor technologies [3,4] were used in an example implementation. Position estimates with uncertainties from three sensor subsystems are loosely coupled with the inertial solution.…”