Optimal Fusion of Sensors.

In complex plants it will often be necessary to use measurements from a large number of sensors in order to get sufficient information about the plant. In general some of the sensor data will be supplementary and some will be complementary. The task of fusioning these measurements is therefore both the task of extracting as much information from the sensors as possible, but also to utilize the redundancy by taking into account the uncertainties and characteristics of the measurements so that the information extracted will be as reliable and complete as possible. How this task should be performed has been investigated in the recent years and a diversity of solutions using as different methods as: neural nets, fuzzy logic, statistical analysis, etc has been suggested. Especially one method has nevertheless proven itself applicable to a wide variety of systems: The Kalman filter. Given a model of the plant, the sensors and the noise affecting the system the Kalman filter is capable of combining the observations in an optimal way (if some prerequisites regarding the linearity of the model and the nature of the noise is satisfied). In the project some generic methods for fusion and management of sensors will be considered. Focusing upon the Kalman filter it will be examined how to:

• Fuse measurements with different sampling rate and quality.
• Use delayed measurements.
• Simultaneously use and calibrate sensors.
• Perform off line autocalibration of sensor and plant. Different sensor types (mainly the camera) will be evaluated with respect to precision and delay.

Furthermore the fusion of data contaminated with noise that violates the prerequisites of the Kalman filter will be examined.

This description is not very up to date I'm afraid. Its from my Ph.D. application from August 95. Hopefully I'll write a better description one of these days.