Deneb Systems / Sky Tracker

Sky Tracker

C++ tracking core, installable as a Python package. Lock onto a target in one call, get position every frame. Tested at 32+ FPS on a Raspberry Pi live video stream, no GPU, no cloud.

Request evaluation Watch demo

Core runtime: C++
CPU only: No GPU
1080p on desktop: 60+ fps
720p on desktop: 180+ fps
Pi live stream: 32+ fps

Core Capabilities

Fast core. Edge-ready profiles.

C++ tracking core with Python and Node.js SDK surfaces. Runs without a GPU - from a developer laptop to a Raspberry Pi 4 on the edge.

EDGE

Runs Without a GPU

Pure CPU tracking - no CUDA, no inference engine, no cloud. The default profile has been tested at 32+ FPS on a Raspberry Pi live video stream.

> default · armv8-a+crc+simd · no OpenCL required

PERFORMANCE

180+ FPS at 720p, 60+ FPS at 1080p

Measured on real footage on a Windows laptop CPU with no GPU. Raspberry Pi live-stream testing reaches 32+ FPS with the default profile.

> 180+ @ 720p · 60+ @ 1080p · CPU only

CORE

C++ Core, pip install

The tracking algorithm is compiled C++17. Install it as a Python wheel - no build step on your side. The same binary powers the CLI, Python, and Node.js surfaces.

> pip install sky-tracker-sdk · pybind11 binding

TARGET MODE

Lock, Track, Recover

Lock one or more targets by bounding box. Stable IDs keep close targets separate; when the image is unclear, a target briefly follows its motion instead of jumping to a neighbour.

> stable IDs · close-target guard · short motion prediction

SDK

Python + Node.js SDK Surfaces

Python bindings for frame-level OpenCV loops. Typed Node.js package for video jobs, CSV parsing, and server workflows. Same telemetry schema across all surfaces.

> same CSV schema: CLI · Python · Node.js

EARLY ACCESS

Free Key, Real Feedback

We're testing on real-world footage. Request a free evaluation key, run it on your target, and tell us what worked and what didn't. A short video is worth more than any metric.

> early tester · free licence · video feedback

defaultRecommended

Best speed/quality balance for most desktop and edge tracking jobs.

correlationIdentity first

Keeps a visually distinctive selected target stable, with less bbox resizing.

adaptiveTighter bbox

Follows target width and height more closely, with additional CPU cost.

SKY TRACKER v2.4

SELECTED-TARGET MODE

0:00 / 0:00

PAUSED

CSV

telemetry

Real Footage

See it in action.

Real sky footage. Drones, birds, and aircraft tracked frame-by-frame with zero pre-training.

MULTI-TRACK

Multi-Target Lock

Visual multi-track demo with color-coded IDs, prediction vectors, and bounding boxes. Production metrics come from labeled gates.

ID SAFETY

Close-Pass Handling

Predicted gates and one-to-one assignment reduce ID stealing during close passes. Ambiguous tracks can coast instead of snapping to a neighbor.

REACQUIRE

Reacquire After Occlusion

After a short coast, size, color, and position cues are scored before new visual evidence is accepted.

Request full demo access

Interactive Demo

Try it Yourself

Mark the target - watch it track in real time

Selected multi-target tracking

Select the targets. Keep the identities.

Sky Tracker runs one coordinated tracking session for the objects you explicitly select. Each target keeps its own identity, motion state, visible box, and telemetry even when paths converge and the visual evidence becomes difficult to separate.

Real multi-target output

01

Select exact targets

Seed each object with its own bounding box. The tracker follows what you chose instead of classifying everything in the scene.

02

Keep stable IDs

Every selected object gets persistent state and its own telemetry stream, so downstream code can address the same target over time.

03

Own each observation

One-to-one observation ownership prevents two tracks from claiming the same visual evidence when objects move close together.

04

Predict before reacquiring

Motion predicts where each box should search next. When evidence is ambiguous, a target can coast briefly instead of jumping to its neighbor.

See the full multi-target workflow

Selection, per-frame updates, identity ownership, and telemetry in one focused demo.

Open multi-target demo

Real World Footage

Every target. Locked.

Birds, drones, and fast-moving projectiles - tracked live from raw video with zero pre-training.

BIRD - PASSERINEAVIAN

01AVIAN

BIRD - PASSERINE

Small bird sample used as visual demo footage. Customer deployments should be scored with labeled clips.

23fpsinput

< 1pxdrift

labelto verify

BIRD - CORVIDAEAVIAN

02AVIAN

BIRD - CORVIDAE

Raven in open-sky flight, useful for qualitative inspection and future customer-style labeling.

30fpsinput

0ID switches

needsgate row

TARGET - FAST OBJECTFAST TARGET

03FAST TARGET

TARGET - FAST OBJECT

Fast object sample for qualitative scale-change demos. Accuracy claims require labeled evaluation.

60fpsinput

SDKdemo

CSVoutput

TARGET - PROJECTILEFAST TARGET

04FAST TARGET

TARGET - PROJECTILE

High-velocity projectile sample for visual review. Use the focused gate or customer labels for metrics.

fastsample

2pxRMS error

0lost frames

Measured Performance

Real footage. Real numbers.

Desktop figures are measured on actual video, not synthetic benchmarks. Raspberry Pi live-stream figures come from real target hardware with the default profile. Identity results describe labeled gates, not every possible scene.

60+fps

Desktop CPU @ 1080p

Windows x64, no GPU

180+fps

Desktop CPU @ 720p

Windows x64, no GPU

32+fps

Raspberry Pi live stream

default profile, no GPU

Fixed-ID switches

labeled two-target close-pass gate

Lock-on call

bbox → tracking in one line

Dependencies

no GPU, no cloud, no inference

SDK Processing Path

Frame Input

OpenCV / numpy / Node video job

Initialize

caller bbox lock

Update

Tracker.update() or trackVideo()

Validate

speed / contrast / size

Reacquire

coast and wide search

Telemetry

state / bbox / confidence

Coast

velocity predict + reacquire

Result

center / bbox / confidence

Platform Status

Windows x64

READY

Linux x86_64

READY

macOS arm64

READY

Raspberry Pi ARM64

READY

Python SDK · OpenCV frame loops

Frame-level control for CV pipelines.

Use Python when you want direct control over frames, bounding boxes, telemetry, and OpenCV integration. It fits notebooks, evaluation scripts, and robotics pipelines.

$pip install sky-tracker-sdk

OpenCV frames
Pass frames directly through the Python binding.
Manual lock
Initialize tracking from your own bbox selector or detector.
Telemetry rows
Read state, confidence, bbox, and frame-level metrics.
Local evaluation
Run labeled footage checks on workstation builds.

Request SDK early access

sky_tracker / track.py

1import cv2, sky_tracker
2
3cap = cv2.VideoCapture("your-footage.mp4")
4fps = cap.get(cv2.CAP_PROP_FPS) or 30.0
5dt  = 1.0 / fps
6
7# Lock onto target in the first frame
8ok, first_frame = cap.read()
9tracker = sky_tracker.Tracker("default")
10tracker.lock(first_frame, bbox=(715, 174, 32, 20))
11
12while cap.isOpened():
13    ok, frame = cap.read()
14    if not ok:
15        break
16    result = tracker.update(frame, dt)
17    print(
18        f"[{result.frame:04d}]  "
19        f"center=({result.cx:.1f}, {result.cy:.1f})  "
20        f"conf={result.confidence:.2f}  "
21        f"state={result.state}"
22    )

Python 3.10+ / Node 18+

sky-tracker v2.4

Pricing

Start free. Ship when it works.

Free 30-day key to test on your own footage. No forms, just email us. Paid plans for when you're ready to ship.

FREE KEYS FOR EARLY TESTERS

START HERE

Evaluation

Free· 30 days

We're in early testing and want real-world feedback. Run it on your footage, tell us what worked - a short video is all we ask.

pip install - no build step
Python SDK + Node.js SDK
.lic key delivered by email
30-day key, no credit card
Share a video clip with us
Tell us what broke or what didn't

Get Free Key

INDIE / RESEARCH

Developer

Under evaluationone-time

perpetual · upgrades 50% off

Perpetual licence for the current version. Single project, no redistribution. Ideal for researchers and hobbyists.

Everything in Evaluation
No expiry after purchase
Single project use
No redistribution
Major version upgrades at 50%

Coming Soon

TEAMS

Commercial

Under evaluation/year

subscription · updates included

Annual subscription for teams embedding the tracker in a product or internal tool. Binary redistribution rights included.

Everything in Developer
All platform updates included
license for 10 projects / month
New SDK surface support
Integration support
Profile tuning available

Discuss Licence

Not sure which plan fits? Email us.

SDK evaluation

Request an evaluation licence.

Attach the SDK request token from your target device. We review each request before issuing a key.

Opens a short form. Evaluation access is granted after review.