Validating the Effectiveness of Adversarial Patches Against Object Detection to Combat Cheats in Online FPS Games

Abstract

With advances in CV (Computer Vision) technology, CV-based cheat tools for FPS games are becoming pervasive. CV-based cheats are more difficult to be detected than cheats that read in-game memory because they use CV to get enemy player’s positions. This paper proposes a method of applying an adversarial patch to each enemy character to avoid YOLOv5-based cheats, and validate the effects of the proposed method on rate of avoiding YOLOv5-based object detection. In addition, this paper evaluates the visibility of pasting an adversarial patch to each enemy through a questionnaire survey.