An Exploration of Human Pose Estimation Based Cheating Tools for FPS Video Game and its Defense Solution

Modern computer vision and AI algorithms have become highly effective in analyzing high-dimensional image and video content for various tasks. Recently, some have exploited the power of computer vision to develop cheating tools for video games, which pose a serious threat to the gaming community and the game industry. Using human pose estimation algorithms, these cheating tools can assist players by automatically targeting and shooting with very high precision and accuracy. Compared to classic cheating methods, these tools are generally much more undetectable as they can mimic real “competent” players. To counter this threat, we propose a machine learning-based approach that leverages the concept of adversarial attacks to generate perturbations that fool such human pose estimation algorithms, preventing cheaters from gaining unfair advantages. In this work, we first implement the video game cheating systems and then we propose and implement our solution. In the end, we use the cheating system to evaluate the efficiency of our proposed algorithms in defending against such cheating tools. Experimental results show that our algorithm can effectively deceive advanced human pose estimation algorithms by adding invisible perturbations to the characters in the video game, maintaining a fair and healthy gaming environment.