YOLO-based Mobile Legends Match Result Parsing

Muhammad Rizqi Nur (a:1:{s:5:"en_US"s:35:"Institut Teknologi Sepuluh Nopember"})
Rarasmaya Indraswari (Institut Teknologi Sepuluh Nopember)

JGGAG (Journal of Games, Game Art, and Gamification)
Vol. 9 No. 1 (2024)

MOBA competitive gaming can benefit from AI advancement. However, data availability is a major issue for Mobile Legends, as opposed to more mature MOBA. In order to obtain high amount of data, it has to be crowdsourced, but it is only viable to collect screenshots. In this paper we propose a framework to automatically parse mobile legends match result screenshots based on YOLO. YOLO is used to locate and classify objects. Text objects are then parsed with OCR. The results are evaluated and compared with older approach using CNN classifiers. The new approach is 25 times faster while achieving the same perfect performance as the old CNN classifier approach.