v2
⚠ Golden Hours protocol

Every Minute
Counts.

AI-powered computer vision to find missing persons during the Golden Hours.

VTX-1 · Meet our drone prototype ↗
SYSTEM STATUS: ONLINE
ACCURACY: 94.2%
CAMERAS: 128 ACTIVE
FRAMES: 48,210
MODEL: VISIONTRACK v2
█
VisionTrack Air / VTX-1

Our drones. A wider view.

From the workbench to the search mission: our team-built prototype brings the drone story into VisionTrack's second generation.

VTX-1 / TEAM-BUILT
VisionTrack team's hand-built quadcopter prototype
VTX-1 · VISIONTRACK AIR UNITACTUAL PROTOTYPE
Built here. Not a stock photo.
01

Above the blind spots

An aerial perspective for rooftops, riverbanks and places street cameras cannot see.

02

One search, two perspectives

Drone imagery joins the camera search, bringing aerial leads into the same review workflow.

03

The decision stays human

Potential match — requires human review. No autonomous pursuit or dispatch.

Prototype shown is ours. Patrol movement and aerial footage are demonstrations, not proof of a live connection or measured flight performance.

Open aerial case view
The problem

The search radius grows with every minute.

The first hours decide most outcomes. Manual review of hundreds of cameras simply can't keep up.

0

cameras per district to review

0×

radius growth from 1h to 6h (walking)

0h

window where leads go cold

0/7

continuous automated scanning

* Illustrative figures

Search radius over time

How it works

From a phone call to a camera frame.

  1. 01
    intake

    A family files a report

    Photos, where and when the person was last seen, what they wore, their age — and any condition like Alzheimer's or autism, because it changes how someone moves.

  2. 02
    match

    It compares more than faces

    CCTV faces are usually blurry, so it also matches clothing, build and how the person walks.

  3. 03
    radius

    It works out where to look

    A seven-year-old walks slower than an adult. Time passed × expected speed = the area worth searching.

  4. 04
    alert

    Operators get a snapshot

    Camera, time, place and a confidence score. Nothing happens automatically.

  5. 05
    human

    A person decides

    An authorized officer confirms or rejects before anyone is sent.

Live demo

Run a search yourself.

Select a profile, start the search, and review the alert like an operator would.

Missing person

Age
7
Clothing
Red hoodie, blue jeans
Last seen
Tahrir Square · 14:20
Condition
Autism — may avoid crowds
Est. speed
2.5 km/h

Camera wall

● REC
Tahrir Sq. — Crosswalk
CAM-014Tahrir Sq. — Crosswalk
Ramses St. — Night
CAM-022Ramses St. — Night
Central Bus Station
CAM-031Central Bus Station
Corniche Sidewalk
CAM-047Corniche Sidewalk
Downtown Junction
CAM-053Downtown Junction
Metro Exit B
CAM-068Metro Exit B

Geo-analysis

Through the model's eyes

Every person is a story.

VisionTrack sees the street the way a detector does — then a human decides.

Street scene with object detection labels
Crowd crossing with person confidence labels
Night sidewalk with detection boxes
Crosswalk with person, bicycle and car detection labels
Aerial intersection with vehicle path lines and signal waves
Pedestrians on a zebra crossing with numbered tracking boxes
Driving scene with 3D boxes around vehicles and sensor path lines
City intersection read by a traffic control system, with vehicle and pedestrian boxes
Overhead crosswalk with person and bicycle detection boxes
Six night camera feeds with person, car and bus labels beside a geo-analysis map
What's inside

Eight things it does.

  • Face + appearance matching
  • Search-radius prediction
  • Real-time alerts
  • Citizen sightings
  • Age progression for older cases
  • Live heat map
  • Several cameras, one track
  • Measured outcomes
Responsible AI

The rules we built it around.

Every alert says “potential match”
The system never claims it found someone. A reviewer does.
Wrong matches are logged
Rejections feed back into the thresholds so the same mistake shows up less.
Tested across ages and backgrounds
We check accuracy separately for children, elderly people and different groups.
Cameras only with permission
Access through official partnerships. Every view is in the audit log.
Data is deleted when the case closes
Encrypted while open, role-based access, removed afterwards.
Abduction mode: metadata first
Telecom data opens only with a judicial order, at metadata level. Message and call content needs a second, supervisor approval.
Full audit trail, minimum data
Every access is logged with officer and time. Only data tied to the case window is pulled.
Plate matches reviewed for bias
Vehicle and plate links are audited for false positives before any checkpoint alert.
Impact

What we’re aiming for in the pilot.

0 min

Time to first match

0%

Alert precision

0%

Reduction in search time

* Target / pilot metrics — not measured results

Roadmap

Three phases.

In progress

Phase 1 — MVP

Photo + location + time → search zone + ranked matches from a sample dataset.

Next

Phase 2 — Pilot

Pilot with real camera partners.

Planned

Phase 3 — National scale

Integration with national authorities and multi-city scaling.

Contact

Running a pilot? Talk to us.

We're looking for camera partners and agencies for Phase 2.

The Team

Two years in a row.

This idea has been developed for the second consecutive year by:

  • Mohamed Nasser

    Project Leader — designed and built the entire website from scratch: the look, the interface, and every page and feature on it. Everything you see here went through his hands.

  • Youssef Salama

    Team leader of the first version — he came up with the original idea and started this project in its first year, leading the team alongside Mohamed Nasser. The idea this whole build is built on is his.

  • Alaa Hani

    A valued member of the VisionTrack team — a name that deserves special recognition in this project's journey. We proudly celebrate her place alongside the team, with sincere appreciation for her contribution to a shared ambition: technology that makes a difference in people's lives.