Computational Complexity Reduction for the Track-to-Track Association Problem

In track-to-track association, the most likely track associations are calculated in order to determine which tracks from various sources can be fused. The most commonly used metric in the various algorithms used for deciding whether two tracks are close enough to be associated is the Mahalanobis distance. However, this distance must be performed between all possible unique pairs of tracks, resulting in a complex quadratic problem. This paper proposes two approaches to speed up and to reduce the computational complexity of the track-to-track association problem. The first one consists in the parallelization of the association process on a multi-core CPU and a GPU to determine the best architecture for the corresponding problem size. The second proposed approach reduces the computational complexity of the association problem by “pre-gating” the pairs of tracks that clearly cannot be associated by using a simpler function, i.e., the squared Euclidean distance. Results show that execution time can be reduced by utilizing multi-core execution on CPU and GPU. The execution time can also be reduced by carefully selecting the pairs of tracks that have a higher probability of being associated. This phenomenon being a direct implication of the reduced number of the computed Mahalanobis distances.