Camera-based perception attacks can manipulate what an automated system detects without accessing its software or hardware. GhostImage research demonstrates how projected light, lens flare and auto-exposure can create or alter objects in camera images.
Articles Tagged: adversarial machine learning
Automotive Cybersecurity: Lessons from Autopilot Research
Research on a 2018 Tesla Autopilot system examined how ECU access, vehicle networks and camera-based perception could affect safety. Its findings offer useful lessons for teams securing connected control systems and machine-learning features.