loading page

Optimization of Anomaly Detection Algorithms for Autonomous Vehicle On-board Sensors: Security Enhancement Methods Based on the Comparison of Theoretical and Measured Values
  • +2
  • Taku Tanaka ,
  • REIKA HOSOMI ,
  • TOMOAKI TANAKA ,
  • Keiko Tanaka ,
  • HIROTAKA CHIKARAISHI
Taku Tanaka
Kyoto Univ

Corresponding Author:[email protected]

Author Profile
REIKA HOSOMI
Author Profile
TOMOAKI TANAKA
Author Profile
Keiko Tanaka
Author Profile
HIROTAKA CHIKARAISHI
Author Profile

Abstract

With the recent advancements in autonomous vehicle technology, there has been a significant surge in the utilization of onboard sensor data. However, concerns regarding sensor anomalies and security issues have concurrently intensified. Malfunctions in sensors can lead to the dissemination of inaccurate data and potential vehicular malfunctions, heightening vulnerabilities in security. This poses a risk to the safety of both drivers and pedestrians and may adversely impact the trustworthiness and adoption rate of autonomous driving technology. In this study, we introduce a novel anomaly detection algorithm, grounded in the operational principles of onboard sensors and surrounding environmental conditions. This algorithm contrasts the actual measurements from the sensors with theoretical estimations to detect anomalies. As a criterion for anomaly detection, instances where the disparity between the estimated and actual values surpasses a specific threshold are identified, with the threshold optimized using the Receiver Operating Characteristic (ROC) curve and Area Under the Curve (AUC). This method aims to achieve high true positive and true negative rates, bolstering the security and reliability of sensors in autonomous vehicles. Additionally, it facilitates performance comparisons across different vehicle types and environments.