On-line metrics for assessment of fused tracking performance

Abstract

We study the possibility of improved tracking of underwater targets based on sensor data from a network of underwater sensors. The improvement by fusion implicitly assumes that the tracking can be assessed in a reliable and objective way. How tracking performance can be assessed is the main issue in this paper. We introduce two categories of performance metrics. The first set, the On-line metrics, have been defined to be independent of a priori knowledge of the target positions. The On-line metrics are intended to be usable in real-time. The second set, the Off-line metrics, is based on comparisons between the estimated tracks and the a, priori known positions (e.g. GPS-positions) of the target. The Off-line metrics are intended for evaluation purposes of the On-line metrics only. The key question is whether the On-line metrics correlate well with the Off-line metrics. In this paper we use experimental data from underwater sensors to assess the tracking performance of underwater targets. Data from acoustic and electric sensor arrays are used either standalone or fused using a Kalman filter. The evaluation is done based on sea-trial data from both sensor-types. The results from applying the On-line and Off-line metrics are discussed. We demonstrate a high correlation between the On-line and Offline metrics estimated from real data. Our conclusion is that the suggested On-line metrics are possible to use as tools for evaluation of tracking performance in real time.