Bringing Player Tracking to Youth Hockey with Computer Vision | Statsyuk Interview

In this session, we chat with Tanner Manett and Evan Cohen, the co-founders of Statsyuk, about how they use computer vision to turn standard hockey footage into detailed, shift-by-shift player tracking data. Detailed player tracking has long required expensive multi-camera rigs and professional budgets. Statsyuk brings it to youth, junior, and collegiate hockey using a single moving camera and a pipeline of 14 computer vision models. The session starts with a live demo of the Statsyuk in action, where we see player tracking, per-shift stats, and team spatial analysis. Tanner and Evan then walk through the computer vision behind it: detecting overlapping players with RF-DETR, tracking through occlusion, reading jersey numbers with Optical Character Recognition (OCR), assigning players to teams, and running it all fast enough for production with ONNX and TensorRT. Chapters: 0:00 Introducing Statsyuk: Computer Vision for Youth Hockey 3:27 Live Demo: Player Tracking, Stats, and Spatial Analysis 9:05 From Detections to Insights for Coaches, Players, and Scouts 17:00 Inside the 14-Model Pipeline: Detection, Segmentation, Tracking, OCR 29:29 Why RF-DETR: NMS-Free Detection for Overlapping Players 36:42 Deploying to Production With ONNX and TensorRT 41:27 Jersey-Number OCR and Temporal Voting for Stable Identity 45:47 Multi-Object Tracking Through Occlusion in a Contact Sport 52:29 Team Assignment With Color Heuristics and a Gaussian Mixture Model 57:10 Roadmap: Puck Tracking, Event Detection, and Lessons From a 14-Model Stack = Resources = Statsyuk: https://statsyuk.ca RF-DETR on Github: http://github.com/roboflow/rf-detr Roboflow Inference: https://inference.roboflow.com Join a future live session: https://roboflow.com/webinar