Data Falsification Attacks on Distributed Multi-Object Tracking Systems

The paper considers the problem of propagation and detection of false information in a distributed sensor network tasked with multi-object tracking. Leveraging the framework of multi-object tracking by means of the Probability Hypothesis Density (PHD) filter, the papers contributes twofold. First, we proposed a new byzantine attack called the overconfident data falsification attack that exploits the knowledge of the data fusion protocol to feed the network with false low uncertainty estimates. Second, we devise a defense strategy within the fusion protocol by introducing time-varying fusion weights that use an inter-agent trust measure based on the Beta reputation system to decide the level of information merging from each neighbouring agent.