Tracking player locations in VALORANT using computer vision

Abstract

Data-driven analytics provide an invaluable competitive edge in traditional sports. With better data availability, they can also be applied in e-sports to generate an advantage and create winning teams. This study proposes a pipeline to automate extraction of location information from in-game screenshots and player-perspective footage using computer vision. Color segmentation and contour detection were used to isolate and classify players according to their involvement in each event. The locations of involved players were extracted from the homography matrices. This information was visualized using heatmaps, with the kill-death ratio (KDR) as the chosen metric. Player locations over time was also visualized using player trails. These analytics are useful in identifying the strengths, weaknesses, and tendencies of a particular team.