Project · software
SoccerNet Plus
Computer vision tools for player detection, tracking, statistics, and a bird’s-eye view of soccer footage.
Motivation
For an undergraduate computer vision course, Aijing Wu and I built an approachable system for exploring soccer footage. We wanted non-specialists to see how computer vision could support player detection, tracking, statistics, and tactical views.
System
The project combined YOLOv3, OpenCV, and DeepSort with an Anvil interface and a Colab-hosted backend. Users could select example clips or provide YouTube footage, then choose between detection and bird’s-eye transformation or player tracking and statistics.
Engineering lessons
We separated model functions from the notebook interface, built video download and clipping logic, and split expensive model calls so users could request only the output they needed. The work sharpened my ability to read technical papers and integrate unfamiliar tools into a coherent system.