Matthew Horsley (Lawrence Livermore National Laboratory)
Keywords: track correlation, Bayesian, bias, marginalization, MCMC
Abstract:
One of the primary goals of Space Situational Awareness is to locate objects in space and characterize their orbital parameters. This is typically performed using a single sensor. The use of multiple sensors offers the potential to improve system performance over what could be achieved by the use of a single sensor through increased visibility, increased accuracy, etc However, to realize these improvements in practice, the association of data collected by the different sensors has to be performed in a reliable manner, with a quantifiable confidence level reported for each data association. Furthermore, sources of error such as track uncertainty and sensor bias have to be taken into account in order for the derived confidence levels to be valid. This paper will describe an approach to compute probabilities of association to support the integration of data collected by multiple sensors on a group of objects. Multi-sensor data association is a fundamental problem in distributed multi-target multi-sensor tracking systems and involves finding the most probable association between object tracks. This is a challenging problem for a number of reasons. Each sensor may only observe a portion of the total number of objects, the object spacing may be small compared to a sensors reported track accuracy, and each sensor may be biased. In addition, the problem space grows exponentially with the number of objects and sensors, making direct enumeration of the possible associations impractical for even modestly sized problems. In this paper, the multi-sensor, multi-target likelihood function will be defined, with sensor bias included in the likelihood function. Sensor bias priors will be introduced and used to marginalize out the sensor bias. This marginalized likelihood will be incorporated into a Markov chain Monte Carlo data association framework and used to compute probabilities of association. In addition, the number of objects is treated as an unknown and probability distributions on this variable will also be produced. Simple problems involving modest numbers of targets and sensors will be described and the multi-sensor track correlation algorithm will be applied, with results presented and discussed.
Date of Conference: September 14-17, 2010
Track: Modeling