VetaDetect: Vehicle tampering detection with closed-loop model ensemble

Piroska Haller, Béla Genge, Fabrizio Forloni, Gianmarco Baldini, Massimo Carriero, Georgios Fontaras · 2022 · 7 citationsRead the paper

Deliberate and unauthorized manipulations of vehicle components, also known as tampering, are nowadays affecting various vehicle functions. The European Commission, alongside several individual countries, estimated a growing number of tampered vehicles. In fact, modern tampering techniques are getting more complex, and are capable of emulating real signals via custom control devices. This paper proposes VetaDetect, a comprehensive methodology for the detection of vehicle tampering. VetaDetect consists of an ensemble of multiple-input single-output (MISO) Auto Regressive Moving Average models (ARX) that are fused together in a closed-loop scenario with the Dempster-Shafer (D-S) theory of evidence. A key feature of the closed-loop detection methodology is that the degree of belief associated to each detector (i.e., ARX model) is adjusted according to the reported belief on tampering. As a result, minority reports on successful tampering detection can influence the outcome of the fusion. Experimental results are based on data collected from a EURO VI D N2 class truck in the Vehicle Emissions Heavy Duty chassis laboratory (VELA) at the Joint Research Centre in the context of a real-world AdBlue (Urea) emulator that is connected to the vehicle. The results demonstrate VetaDetect’s ability to detect real-world and more intelligent tamperers, and its superior performance when compared to state of the art techniques such as the ones leveraging Long Short-Term Memory (LSTM)-autoencoders. Accordingly, VetaDetect shows better accuracy to detect real-world tampering (98%) in comparison to most recent works (54%), while being up to 9 times less computationally intensive.

1 idea Seedlabs derived from this research