Skip to main content

Kestrel

Crowd and traffic analytics that runs without a GPU

kestrel-vision.vercel.app
Kestrel interface
Real-time tracking without a GPU
CPU only

Overview

What Kestrel does.

Counting people and vehicles from a video feed in real time, on ordinary CPU hardware. Detection runs through ONNX Runtime rather than a training framework, which is what makes commodity hardware sufficient — and keeps deployment cost proportionate to the problem.

  • Detection and tracking with line crossing counts

  • Zone occupancy and heatmaps over time

  • Results streamed live over WebSockets

  • CPU-only — no GPU in the deployment footprint

Keep exploring

More in Vision & Edge AI

  • Live
    Built withMobileNetV2ONNX Runtime WebTypeScriptPWA

    LeafWise

    Crop disease detection that works with no signal

  • Live
    Built withMediaPipeTypeScriptWebGLReact

    Kinetix

    Movement analysis in the browser, on-device

Let's talk

Have a problem worth solving?

Tell us what you're building. We'll help you scope it, build it, and ship it.