Performance Evaluation of Multi-platform Distributed Data Fusion Methods for Multi-target Tracking

Abstract

While sensor accuracy cannot be increased beyond a limit, the performance of target tracking algorithms can be greatly enhanced by employing multiple sensors with overlapping coverage regions. An efficient data fusion algorithm is the key to this improvement. In the current work we discuss in detail three distributed data fusion algorithms, namely, track-to-track fusion, tracklet fusion, and associated measurement fusion. These algorithms fuse the information from the sensors at different stages of processing. Their performances are compared with that of centralized fusion, which fuses the unprocessed information (measurements) from the sensors. The sensors measurements are considered asynchronous, though the fusion times are synchronized on all sensors. The scenarios used for comparison contain multiple targets with close and crossing trajectories, involving the data association assessment as well. The fusion performance metrics used in this evaluation are also explained. They are estimated using the Monte Carlo method.