Vision analytics · Cameras · Drones · Retail · Safety

Detect, track, count Live, from any camera

We turn live RTSP camera streams and 4K 30 fps drone video into stable tracks, live counts, and structured events—quantized to ONNX int8 for edge hardware or served centrally on Triton.

Problem
Hours of camera and drone video that nobody has time to watch.
Proven
Multi-camera tracking and re-ID, ONNX int8 on edge devices.
replay · intersection.mp4 · D-FINE-S
REPLAY00:00.0
D-FINE-S · offline
00:12.0

Offline D-FINE-S tracks on this clip, no hand-placed boxes. Tap a box to inspect it.

Architecture

Built for production Not a slide deck

What you get

Concrete outputs Measured on your data

01

Stable track IDs

D-FINE and YOLO detection plus SAM segmentation feed an IoU + Kalman motion tracker, OSNet appearance re-identification, and detector re-prompting through occlusion.

02

Counts and events, live

Structured detections and track events flow through FastAPI into ClickHouse and a Next.js live dashboard for real-time monitoring and historical queries.

03

Edge-sized models

Models are quantized and exported to ONNX int8 for edge deployments or hosted on Triton GPU servers when centralized inference fits your site.

How we deliver

Small team Short loops

  1. 01

    Start from your real data

    A week with your documents, cameras or logs. We agree on one number that means “working” and freeze a test set before we train anything.

  2. 02

    Prototype behind a gate

    Baselines first, then models and agents. Every change runs against the frozen set, so progress is a chart, not an opinion.

  3. 03

    Harden for production

    Quantize, serve, trace and cost it. Retraining, rollback and monitoring are built in, not bolted on.

  4. 04

    Hand over, stay close

    Your team gets the code, the decision records and a walkthrough. We stay on call while it settles.

FAQ

Common questions Direct answers

Can the system work with our existing RTSP cameras and drone footage?
Yes. The pipeline ingests standard RTSP camera feeds through OpenCV as well as 4K 30 fps drone video, without requiring proprietary camera hardware.
How do you keep track IDs consistent across occlusions and multiple cameras?
We pair D-FINE or YOLO detectors with an IoU and Kalman filter tracker, detector re-prompting, and OSNet appearance re-identification so targets stay linked across occlusions and camera views.
Do models have to run in the cloud, or can they run on edge devices?
Both are supported. We export quantized ONNX int8 builds for edge hardware where bandwidth or latency matters, and deploy Triton GPU servers when centralized multi-stream inference makes more sense.
How do you verify tracking and counting accuracy before rollout?
We spend the first week on your real camera footage, agree on one target metric, and freeze a test set. Every model, tracker, and quantization change is measured against that frozen set before production hardening.

Have a problem AI should solve?

Tell us what the data looks like and what “working” would mean for you. One of the founders will reply.