Optimal Positioning for Your Champion: Data-Driven Insights from League of Legends Player Trends

Introduction

In this post, I explain a project aimed at helping new League of Legends players like my friend, who had trouble knowing where to be on the map. We looked into how experienced players position themselves and how that affects their chances of winning.

How to solve this problem ?

To know where a player should be on the map, we first need to extract the map from a screenshot of the game using an AI. From there, we extract key information such as player locations, towers, and objectives. Finally by tracking player movements over time, we can create a heatmap to visualize optimal positioning.

Detect and extract the map

To detect and extract the map, we will use an AI that identifies and draws bounding boxes on our images. To achieve this, we first need to build a dataset showcasing different player setups, as the map is customizable.

After labeling approximately 30 images using Label Studio, I trained an initial Fast R-CNN model to draw bounding boxes around the map.

Here are some predictions from my test set. As you can see, the model performs well, as it is able to detect the map even when there’s a camera on it, when the YouTube progress bar is present, or when the map is cropped and incomplete.

Detect and extract player locations, towers, and objectives

We will extract each map and create a new dataset of map crops to labellised players towers and objectives on them.