Decentralized tracking with feedback, adaptive sample rate and IMM
Erik Karlsson · 2000
This report compares filter architectures for target tracking with multiple sensors. Two decentralized architectures are implemented and evaluated in simulations against a central tracking filter. Data from an array radar and an infrared sensor are used. The radar has adaptive sample rate, i.e. the radar is only used when necessary. The central filter is an interactive multiple models filter (IMM) with three models. These models are extended Kalman filters. The decentralized architectures have two sensor filters and one central filter that fuses the tracks from the sensor filters. Feedback to the sensor filters is used. The sensor filters are also IMM-filters. The difference between the two decentralized architectures is the amount of information that is communicated between the filters in the architecture. The filter architectures are evaluated with respect to their accuracy and ability to perform adaptive sampling. The conclusion is that the decentralized filter that communicates the most is more accurate and needs fewer radar measurements than the filter that communicates less. The central filter is the best tracker as expected and the best decentralized filter is almost equally good.