Adaptive semi-greedy search for multidimensional track assignment
Samuel Shapero, Hunter D. Hughes, Peter B. Tuuk · International Conference on Information Fusion · 2016
Robust multitarget, multisensor fusion algorithms rely on optimal or near-optimal assignment algorithms at their core. Greedy and semi-greedy search algorithms represent a fast and effective solution to multi-dimensional assignment problems, at the cost of often producing sub-optimal results in challenging cases. The Adaptive Semi-Greedy Search (ASGS) presented here is an improvement on older semi-greedy search methods, and is more likely to find the optimal solution without sacrificing computational speed. ASGS outperforms Semi Greedy Track Selection (SGTS) in a wide range of statistical tests, where it was consistently more likely to find the optimal solution to 6 dimensional assignment problems, as if the RMS measurement error had been reduced by 25%. The two algorithms are also compared in several challenging multitarget, multisensor tracking scenarios, including maneuvering formations and crossing targets, where a Multiple Hypothesis Tracker (MHT) using the ASGS significantly reduces the RMS position error of the tracks (p < 0.001 for each scenario). The ASGS effectively increases the range of scenarios in which semi-greedy assignment algorithms can be reliably employed, especially valuable for real time applications with large number of targets and sensors.