Temporal Clustering in the Multi-Target Tracking Environment
Abstract
In multi-target tracking problems such as those found in high-energy particle physics, fluid mechanics, and ballistic missile defense, the common objective is to separate the data into observations associated with individual targets and to use this data to estimate the targets' trajectories. In defense related applications, it is necessary to have algorithms which are computationally efficient, robust, and minimize data storage requirements. Recently developed approaches in the field of multi-target tracking, however, have been shown to have significant computational disadvantages. In this study, non-hierarchical clustering methods are combined with computationally efficient algorithms such as those used to solve assignment and quadratic programming problems to provide an integrated procedure which is computationally efficient, minimizes data storage requirements, and gives a reasonable estimate of the number of targets. Combined with a sequential estimation filter such as the extended Kalman filter, the procedure can provide estimates of a target's state and state covariance after three observations and continuously maintain updated target state estimates in real time. Empirical results based on 100 targets in ballistic trajectories have demonstrated this method's effectiveness by properly clustering data with four measurement attributes (range, range rate, azimuth, and elevation) in over 98 percent of the cases. Theses.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1988
- Accession Number
- ADA197153
Entities
People
- Thomas S. Kelso
Organizations
- Air Force Institute of Technology