Classical methods for multiple target tracking (such as the Joint Probabilistic Data Association (JPDA) Filter or the Nearest Neighbor methods) rely on the suboptimal decomposition of the problem into three steps: track initiation, maintenance and deletion. This paper presents an alternative solution that solves the tracking problem in an integrated way, through a global optimization. This problem is known to be a complex combinatorial one, for which neural models are particularly interesting. Previous works, based on the Hopfield model, proposed neural solutions that are drastically sensitive to the choice of some parameters, specially the size of the network and the weights in the cost function to be minimized. This kind of network often converges towards unacceptable solutions (typically: several plots were associated with the same track, or vice versa). We propose a new neural solution based on a recursive mode, where the constraints are taken into account in introducing competition between the neurons. The network optimizes an objective function that is a measure of the global quality of the tracking, computed as the sum of the track qualities at a given antenna turn. We finally present some simulation results for multiple target monosensor tracking that enable a comparison with classical techniques.
Neural network for track initiation and maintenance
1997
10 Seiten, 17 Quellen
Conference paper
English
Track maintenance - Track monitoring on the TGV network
Online Contents | 2001
IuD Bahn | 2011
|Engineering Index Backfile | 1915
|Strategies for Automatic Track Initiation
NTIS | 1979
|Springer Verlag | 2013
|