ISIS project: Sensor Fusion

This project is carried out by Division of Automatic Control and Communication Systems in cooperation with SAAB Aerospace, SAAB Bofors Dynamics and NIRA Dynamics.

Selected publications

F. Gustafsson, F. Gunnarsson, N. Bergman, U. Forssell, J. Jansson, R. Karlsson, and P-J. Nordlund. Particle filters for positioning, navigation and tracking. IEEE Transactions on Signal Processing, Vol. 50, Nr. 2, 2002. pdf-file.
N. Persson, F. Gustafsson and M. Drevö. Indirect tire pressure monitoring using sensor fusion. In SAE 2002, Detroit, 2002. Report number 2002-01-1250. pdf-file.
P-J. Nordlund and F. Gustafsson. Recursive estimation of three-dimensional aircraft position using terrain-aided positioning. Proc. of ICASSP, Orlando, May, 2002. pdf-file
P-J. Nordlund, F. Gunnarsson and F. Gustafsson. Particle filters for positioning in wireless networks. Proc. of EUSIPCO (invited paper), Toulouse, France, September, 2002. pdf-file.
Urban Forssell, Peter Hall, Stefan Ahlqvist, Fredrik Gustafsson. Novel Map-Aided Positioning System. Proc. of FISITA, Helsinki, Finland, 2002. pdf-file.
Positioning using time-difference of arrival measurements. Fredrik Gustafsson and Fredrik Gunnarsson. ICASSP, Hongkong, PRC, 2003. pdf-file or ps-file.
Particle filter and Cramer-Rao lower bound for underwater navigation. Rickard Karlsson, Fredrik Gustafsson and Tobias Karlsson. ICASSP, Hongkong, PRC, 2003. pdf-file or ps-file.

Highlights 2002

Three licentiate theses were presented:

Rickard Karlsson. Simulation Based Methods for Target Tracking. LIU-TEK-LIC-2002:930, Dept. of Electrical Engineering, Linköping University, Sweden, 2002.
Per-Johan Nordlund. Sequential Monte Carlo filters and Integrated Navigation. LIU-TEK-LIC-2002:945, Dept. of Electrical Engineering, Linköping University, Sweden, 2002.
Niclas Persson. Event Based Sampling with Application to Spectral Estimation. LIU-TEK-LIC-2002:981, Dept. of Electrical Engineering, Linköping University, Sweden, 2002.

A project survey paper appeared, describing most of the projects above and a general framework for the particle filter.

Particle filters for positioning, navigation and tracking.

pdf-file

The author of the master thesis

Hall, Peter. "A Bayesian approach to Map-Aided Vehicle Positioning". NIRA Dynamics, LiTH-ISY-EX-3102, 2001. received an award (nonexisting link removed). The project was successfully continued in a student project and another master thesis. The result is a complete navigator including voice guidance implemented on a Compaq iPAQ handheld computer. This is probably the world's first demonstrator of a real application based on a particle filter.

A new company, NIRA Dynamics, entered ISIS and the sensor fusion project.

A workshop was organized on particle filters for navigation, including a plenary talk and three master thesis presentations at our three industrial partners. The workshop was attended by about 50 people.

Overview

Mobile phone positioning from time-difference of arrival measurements

The research in the area of sensor fusion is focused on target tracking, navigation and positioning applications. The common denominator here is that these problems can be written as non-linear state space models with non-Gaussian noise:

x(t+1) = f(x(t)) + w(t)
y(t)=h(x(t)) + e(t)

where the state vector contains position, velocity and other dynamical states, and the measurements come form sensors as gyroscopes, accelerometers, GPS, radar, etc. The filtering problem is to estimate the states given the measurements. The classical approach is to linearize the state space model, to assume Gaussian noise and then applying the Kalman filter. The particle filter provides a general algorithm for approximating a posteriori distirbution of the states with arbitrary accuracy. The framework is particular suitable for sensor fusion, where sensor information of different kind is mixed with e.g. information from digital maps.

We have the last few years developed efficient and robust particle filter algorithms to approximate the a posteriori distribution of the state vector, and applied the algorithms to a number of applications:

Terrain navigation. In this application an aircraft position is autonomously determined by fusing measurements from an inertial navigation system, a digital map and a radar altimeter. By measuring the terrain height variations along the aircraft flight-path and comparing these with a digital terrain map, a position estimate of the aircraft is obtained. The comparison between the map and the measurements is a nonlinear estimation problem where unconventional and conceptually different sources of information are fused together. Research as been focused on finding a reliable and effective algorithm for this position determination. See the paper
Underwater navigation. The terrain navigation project has been extended to underwater navigation using sonar depth measurements. An underwater terrain map replaces the terrain map above, and similar inertial sensors can be used. The Cramer-Rao lower bound on the estimation accuracy has been derived, and a simple interpretation is that
See the image above, particle filter master theses overview and the conference paper
Map aided positioning for vehicles. An approach similar to terrain navigation can be applied to cars, when a street map is available. The velocity of the wheels is compared to the street map, and after a few turns the correct position is usually found. See the particle filter master theses overview, the awarded master thesis summary (nonexisting link removed) and the conference paper
Target tracking. Target tracking algorithms for torpedos hunting ships using bearings-only measurement and for aircraft hunting ships using map information are presented in Rickard Karlsson's licentiate thesis. Also the data association problem has been studied.
Mobile terminal positioning. A survey of tracking algorithms for mobile terminals is given in
Positioning using time-difference of arrival measurements based on both a stochastic gradient algorithm and particle filter is described in