Computer Vision Engineer — Sports Broadcast Analytics

Leila Ahmadi

Freelance · Mid-level · Sydney

This role is no longer accepting applications

Similar open roles you can apply to now:

About the role

We build sports analytics products for broadcast and team performance applications in the APAC region. Our primary clients are cricket boards, rugby unions, and football leagues across Australia, New Zealand, India, and Southeast Asia. Our current product suite covers ball-tracking, player position heat maps, and automated highlight detection from broadcast footage. We are commissioning a ten-week project to extend our football analysis capabilities: specifically, we want to build an automated player action classification system that can identify and label discrete game events — passes, shots, tackles, dribbles, headers — from broadcast video without requiring manual tagging. We have three seasons of annotated match footage from two A-League clubs that have consented to use in model development. The goal is a prototype system that takes a raw match video and outputs a timestamped event log with action labels and player identifiers. If the prototype meets our accuracy targets, we will integrate it into our broadcast analytics pipeline and we would expect to continue with you on ongoing development.

Key Deliverables

List the expected deliverables for this project.

Design and train a player action classification system for broadcast football video using our annotated A-League dataset
Implement a multi-object tracking component to maintain player identities across camera cuts and rapid movement sequences
Evaluate model performance against agreed accuracy targets with clear breakdown by action type and scenario
Optimise inference speed for near-real-time processing on our existing GPU serving infrastructure
Deliver a structured technical handover including documented architecture, training pipeline, evaluation results, and recommendations for the next development phase

Requirements

Technical stack needed for this mission.

4+ years of computer vision engineering with experience in video understanding, action recognition, or sports analytics
PyTorch for developing and training video classification or temporal action detection models
Python for data pipeline development, annotation tooling integration, and model evaluation
Computer Vision — beyond static image classification: optical flow, temporal modelling, multi-object tracking, and the challenges of broadcast footage with varying camera angles and occlusions
Docker for reproducible training and inference environments that our engineering team can build on
Scikit-learn for baseline model evaluation and performance metric computation
Familiarity with sports video analysis or action recognition literature is a genuine advantage — you should be able to evaluate existing approaches and explain why you are choosing the one you are implementing