Bayesian Inference for Autonomous Personal Localisation Indoors

The principal concept of navigation is to start from a known (initial) position and to ensure a continued and reliable localisation of the user during his/her movement. The initial position of the trajectory is usually obtained via GPS or defined by the user. Consider a pedestrian navigation system which contains a set of inertial sensors, connected with a map database. In the urban environment and indoors the localisation depends entirely on the measurements from the inertial sensors. The trajectory is defined in a local coordinate system and with an arbitrary orientation. The problem to solve is to determine the users location using the map database and inertial measurements of the navigation system. The idea behind our approach is to find the location and orientation of the trajectory and thus the users location. The proposed solution associates the users trajectory with the map database applying statistical methods in combination with map-matching. Similar geometric forms must be identified in both the trajectory and the link-node model. The trajectory, defined by a set of consecutive points, is transformed to a set of lines thanks to a dedicated motion model. In this research we propose a solution based on statistical methods where the history of the route and actual measurements are treated at the same time. The determination of the absolute position is entirely represented by its probability density function (PDF) in the frame of Bayesian inference. Following this approach the posterior estimation of the users location can be calculated using prior information and actual measurements. Because of the non-linear nature of the estimation problem, non-linear filtering techniques like particle filters (Sequential Monte Carlo methods) are applied.

Bayesian Approach for Indoor Pedestrian Localisation

Michel Bierlaire, Bertrand Merminod

The principal concept of navigation is to start from a known (initial) position and to ensure a continued and reliable localisation of the user during his/her movement. The initial position of the trajectory is usually obtained via GPS or defined by the user. Consider a pedestrian navigation system which contains a GPS receiver and a set of inertial sensors, connected with a map database. In the urban environment and indoors the localisation depends entirely on the measurements from the inertial sensors. The trajectory is defined in a local coordinate system and with an arbitrary orientation. The problem to solve is to determine the users location using the map database and inertial measurements of the navigation system. The idea behind our approach is to find the location and orientation of the trajectory and thus the users location. The proposed solution associates the users trajectory with the map database applying statistical methods in combination with map-matching. Similar geometric forms must be identified in both the trajectory and the link-node model. The trajectory, defined by a set of consecutive points, is transformed to a set of lines thanks to a dedicated motion model. In this research we propose a solution based on statistical methods where the history of the route and actual measurements are treated at the same time. The determination of the absolute position is entirely represented by its probability density function (PDF) in the frame of Bayesian inference. Following this approach the posterior estimation of the users location can be calculated using prior information and actual measurements. Because of the non-linear nature of the estimation problem, non-linear filtering techniques like particle filters (Sequential Monte Carlo methods) are applied.

2006

Map-matching for pedestrians via Bayesian inference

Michel Bierlaire, Bertrand Merminod

A navigation process is to start from a known (initial) position and to ensure a continued localisation of the user during the movement. Consider a pedestrian navigation system which contains a GPS receiver and a set of inertial sensors connected with the map database. The problem to solve is to determine the users location using the map database and measurements of the inertial sensors. Indoors the position of each step is determined as a function of the previous position and inertial measurements. Thus the trajectory is defined in a local coordinate system and with an arbitrary orientation. A dedicated motion model transforms the trajectory from set of consecutive points to a polygon. Then we must associate similar details from both data sources, the modified trajectory and the link-node model. The trajectory can be considered as the history of the route and its last point as the actual position of the user. In this research we propose a solution based on statistical methods and map-matching. The determination of the absolute position is entirely represented by its probability density function (PDF) in the frame of Bayesian inference. Following this approach the posterior estimation of the users position can be calculated using prior information and actual measurements. Because of the non-linear nature of the estimation problem, non-linear filtering techniques like particle filters (sequential Monte Carlo methods) are applied.

2006

Map-based Autonomous Personal Localization and Tracking

Michel Bierlaire, Bertrand Merminod

An autonomous localisation and tracking method is a method independent from the reception of external data. In our approach we ignore methods like GPS- and WiFi- positioning and we focus on the use of inertial navigation system (INS) carried by the person and connected to a map database. The walking person is considered as a dynamic system, whose movements are measured by the INS. His trajectory is modified with respect to a dedicated motion model. Users location is estimated in the frame of Bayesian inference and is based on the association of the trajectory to the map database, a technique known as map- matching. Because of the non-linear nature of the estimation problem, non-inear filtering techniques like particle filters (Sequential Monte Carlo methods) are applied. In tracking mode simple geometric constrains will be observed in order to associate every point of the trajectory to element of the map database. In parallel the localisation mode rest active in order to keep the knowledge on the history of measurements. The algorithm is tested on the campus of EPFL; the process of localisation is entirely autonomous and gives promising results. That method of localisation can be applied to many pedestrian navigation tasks, in particular for the needs of the fire-brigades and security services.

2007