Character Detection in First Person Shooter Game Scenes Using YOLO-v5 and YOLO-v7 Networks

Game character detection methods have been the basis in the design of artificial intelligence in video games. With real-time character detection methods based on in-game views, the game AIs can become more realistic and life-like compared to traditional information-gathering methods on memory-level. This is particularly the case in first person shooting (FPS) games, where it is extremely important to extract information directly from the in-game view the players have. However, there are difficulties in in-game view based character detections, such as viewpoint variation, ambient illumination and occlusion. The purpose of this research is to compare the performance of two famous versions of a CNN-based model named You-Only-Look-once (YOLO), including the most popular version named YOLOv5 and the state-of-the-art version, YOLO-v7, to find out the currently optimal method to be used in the further development of game AIs.