Technology

Built on Science

Our AI pipeline transforms standard video into deep athletic insight in real-time.

1

Input

Computer Vision & Pose Estimation

Our models process standard video frames to detect and track athletes in real-time. Using state-of-the-art pose estimation, we extract precise 2D joint positions at up to 60fps—even in complex multi-person scenes with occlusions.

  • Markerless tracking from any camera angle
  • Multi-person detection and tracking
  • Works with broadcast, phone, and training footage

Pose Estimation Pipeline

2

Processing

3D Reconstruction & Signal Processing

2D joint positions are lifted into 3D space using our proprietary depth estimation models. Advanced signal processing algorithms then extract biomechanical metrics—angular velocity, rotational dynamics, trajectory analysis, and force estimation.

  • 2D-to-3D skeletal lifting
  • Biomechanical model fitting
  • Sport-specific metric extraction

3D Reconstruction Engine

3

Delivery

Real-time Delivery & Integration

Processed data is delivered in real-time through our API and rendering pipeline. For broadcasts, graphics are generated and composited live. For athletes, insights appear instantly in the mobile app with full historical context.

  • Sub-second end-to-end latency
  • Standard broadcast protocol support (NDI, SDI)
  • API-first architecture for custom integrations

Real-time Delivery

Recognition

Grounded in Research

Our work has been featured in leading scientific and sports publications for advancing the intersection of AI and athletic performance.

MIT
Scientific American
Associated Press
NBC Sports
Sports Business Journal
Apple App Store

Specs

By the Numbers

60fps

Pose Detection

<1s

End-to-end Latency

33+

Body Keypoints

5+

Sports Supported