Generic Distributed Target Tracking Algorithms in Sensor Networks with Finite Set Statistics

本研究计划将研究和开发新的分布式多目标多源检测（DMMD）和跟踪算法的传感器网络与有限的通信资源。目前的方法DMMD广义分布式数据融合（DDF）算法，将它们与多假设跟踪（MHT）算法相结合。然而，MHT中固有的近似会导致跟踪性能的不可接受的退化。为了克服这个困难，我们提出了一个新的DMMD算法，建立在有限集统计（FISST）和指数混合密度（EMD）。FISST提供了一个严格的和数值易处理的模型，统一的多目标多传感器检测，分类和估计的问题。EMD是一种次优的融合估计算法时，他们的边缘分布是已知的，但他们的联合分布是不。它可以被用来融合估计融合网络的网络拓扑是任意的，未知的和时变的。将有两个主要成果，从这个研究计划：首先，我们将创建一个非常普遍和通用的数学框架内，一系列的非线性滤波算法可以部署。其次，我们将开发的实现，我们相信，将显示显着的优势，现有的方法，在他们的能力，以处理高误报率和数据关联的模糊性。我们还将努力提高计算效率和实用性。成功地扩展到分布式环境可能具有广泛的适用性，由于其简单的实现和低复杂性。该计划是在响应的EPSRC-DSTL呼叫的检测要求和挑战号13：``开发通用算法的分布式信号融合网络的传感器。''

This research programme will investigate and develop new distributed multi-target multi-source detection (DMMD) and tracking algorithms for sensor networks with constrained communication resources. Current approaches to DMMD have generalised distributed data fusion (DDF) algorithms by combining them with multiple hypothesis tracking (MHT) algorithms. However, the approximations inherent in MHT can lead to an unacceptable degradation in tracking performance. To overcome this difficulty, we propose to develop a new DMMD algorithm that builds upon Finite Set Statistics (FISST) and Exponential Mixture Densities (EMD). FISST provides a rigorous and numerical tractable model that unifies the problems of multi-object multi-sensor detection, classification and estimation. EMD is a suboptimal algorithm for fusing estimates when their marginal distributions are known but their joint distribution is not. It can be used to fuse estimates in fusion networks where the network topology is arbitrary, unknown and time varying.There will be two main outcomes from this research programme:First, we shall create an extremely general and generic mathematical framework within which a range of non-linear filtering algorithms can be deployed. Second, we shall develop implementations that, we believe, will show significant advantages over existing approaches in their ability to deal with high false alarm rates and data association ambiguity. We shall also strive for computational efficiency and practical applicability. The successful extension to distributed environments could have widespread applicability due to their simplicity to implement and low complexity.This programme is in response to the Detection requirement and Challenge Number 13 of the EPSRC-DSTL call: ``To develop general algorithms for distributed signal fusion in a network of sensors.''