Proposed a distributed mean-field-type filter for signal processing.
The combination of observed data and dynamical
models of mean-field type over networked systems is a challenging problem because of non-linearity, high dimensionality and partial observations. In many networked systems, the effective extraction and utilization of the information contained in observed data should be accomplished by exploiting the availability of accurate predictive, proactive tools of mean-field type dynamical systems. Incorporating observed big data into dynamical models of mean-field type has two problems. One is the curse of dimensionality, and the other is the control of error accumulation. This paper presents a distributed meanfield filter (DMF) framework for large scale networked systems. The proposed filter exploits the topology of the network and
decomposes it into highly independent components with respect to the marginal mean-field correlations. The upper bound of global filtering error can be estimated using mean-field-type
game theory. Numerical experiments in two object tracking scenarios are carried out to illustrate the performance of our algorithm. Evaluation results show that DMF significantly
outperforms the existing filtering algorithms.
The main contributions of this paper are the following:
1. Proposed a distributed mean-field-type filter (DMF) for traffic networks. It exploits the topology of the network and decomposes it with respect to the marginal correlations;
2. Provided a generic methodology to estimate the filtering distribution in four steps: sampling, prediction, decomposition and correction. This methodology can be applied to other high-dimensional networked systems with mean-field-type dynamic models and noisy observations; 3. Proved the filtering error can be bounded by a linear term with respect to the number of decomposed zones;
4. Implemented our algorithm on two object tracking scenarios and evaluate the performance. The experimental results demonstrated the advantage of our algorithm in comparison with the existing filters.
--- Aircraft Tracking
This video shows the distributed mean-field filter applied in an application that involves estimating the aircraft positions through a model for RAdio Detection And Ranging (RADAR) measurements. There are three independent zones where the green mountain areas are available for RADAR detection, while the blue sea area has no signal. Our filtering algorithm is used to estimate the possible aircraft locations based on the partial and noisy sequential observation.
This work has been published at IEEE International Conference on Data Science and Systems (HPCC-SmartCity-DSS) in 2016.