A Kinect for everything, minus the Microsoft lock-in

What it does Nuitrack is commercial skeletal-tracking middleware that ingests depth-sensor streams and outputs 3D body poses, hand gestures, facial analysis, and scene segmentation. It targets embedded boards like Raspberry Pi 4 up to multi-sensor analytical rigs, shipping runtime packages for Windows, Linux, Android, and iOS. The SDK wraps the core engine in C++, C#, and Python bindings, plus plug-ins for Unity, Unreal Engine 5, TouchDesigner, and Docker.

The interesting bit The hardware compatibility list is deliberately nostalgic and practical: Orbbec, Femto Bolt/Mega, Intel RealSense, Kinect v1/v2/Azure, and even the aging Asus Xtion are all supported. It offers two tracking backends—a lightweight “classical” engine for CPU-constrained hardware and a deep-learning “AI” engine for complex poses—letting developers trade power for precision depending on the deployment target.

Key highlights

• Dual tracking engines: a fast classical pipeline and an AI-based pipeline for harder poses.
• Broad sensor and platform support, from ARMv7 embedded devices to x64 desktops and iOS.
• Native Unreal Engine 5 MetaHuman integration and Unity plug-ins for immersive applications.
• Embeddable runtime that can be shipped inside your application without a separate end-user install.
• Academic track record: cited in robotics, gait-analysis, and human–robot interaction research.

Caveats

• Several advertised features and bindings are explicitly labeled beta, including Python support, iOS, and the Nuitrack Holistic multi-sensor fusion system.
• The core runtime is commercial software requiring license purchase; this repository distributes the SDK, bindings, examples, and documentation rather than the full engine source.
• Facial analysis and age/emotion detection are described as “basic,” so expectations for advanced computer-vision tasks should be modest.

Verdict Worth evaluating if you are building gesture-driven UIs, fitness apps, or AR/VR full-body tracking and need to support a mix of depth sensors without rewriting your pipeline for each one. Skip it if you are looking for a fully open-source, license-free tracking stack.