Sensor Compromise Detection in Multiple-Target Tracking Systems.

Juan-Pablo Ramirez-Paredes,Jess Curtis,Emily Doucette,Victor Ayala-Ramirez

doi:10.3390/s18020638

Juan-Pablo Ramirez-Paredes, Jess Curtis + Show 2 more

Open Access

https://doi.org/10.3390/s18020638

Copy DOI

Abstract

Tracking multiple targets using a single estimator is a problem that is commonly approached within a trusted framework. There are many weaknesses that an adversary can exploit if it gains control over the sensors. Because the number of targets that the estimator has to track is not known with anticipation, an adversary could cause a loss of information or a degradation in the tracking precision. Other concerns include the introduction of false targets, which would result in a waste of computational and material resources, depending on the application. In this work, we study the problem of detecting compromised or faulty sensors in a multiple-target tracker, starting with the single-sensor case and then considering the multiple-sensor scenario. We propose an algorithm to detect a variety of attacks in the multiple-sensor case, via the application of finite set statistics (FISST), one-class classifiers and hypothesis testing using nonparametric techniques.

Highlights

Including information from multiple cooperative sensors in a target tracking scenario can increase performance over using a single sensor or uncooperative sensors
The main difference between the two approaches is that ours uses specific transition probabilities for each sensor, while the support vector machines (SVMs) algorithm is trained on pure data and does not include information about the likelihood of the reliability of each sensor
In order to illustrate the effectiveness of both the Bayesian networks (BNs) and the SVM approaches under some attacks, we show in detail the results of three simulated attacks

Summary

Introduction

Including information from multiple cooperative sensors in a target tracking scenario can increase performance over using a single sensor or uncooperative sensors. As the number of sensors in a network increases, as does the number of potential vulnerabilities. This concern has been the subject of many studies by the cyberphysical systems community [1,2,3]. We consider some state space Es for the targets. The multiple-target state at some time instant k, assuming M(k ) targets are present, is given by. Eo is the space of all possible observations, and the N (k) multiple-target measurements at time k are given by. The RFS that represents the multiple-target state at time k is

Methods

Results

Discussion

Conclusion