HatchWatch

Map Page
Camera Page
Live Cameras Page
Analytics Page
Settings Page
Camera Pole Mount
Camera Pole Mount

Inspiration

On the beaches around Port Aransas, sea turtle nests are found the hard way: search teams walk miles of sand at dawn, hoping to stumble on a fresh crawl before tides, traffic, or predators erase it. Whether a nest gets protected comes down largely to chance. We asked: what if the beach could tell the patrol where to look?

What it does

HatchWatch watches the beach with pairs of cameras and flags turtle digging activity: it never claims to confirm a nest. Each alert reads exactly "Digging activity detected; nesting unresolved." and drops a pin on a live map so a volunteer can walk straight to it, log what they find, and mark the site protected.

A live dashboard (map, alerts, nest log, analytics, patrol roles) shows every site and its status.
Any phone can become a camera: open a page, tap Start streaming, and an admin assigns it to a spot on the map.
A side-by-side motion view shows a live feed next to what is moving in it.

How we built it

We had no permit to film nesting turtles, so we built the beach first: a 3D simulator (TypeScript and Three.js) with sand made of individual grains, where turtles crawl up, dig, and leave while two cameras record from opposite ends.

A Python and OpenCV pipeline watches those videos blind. It only sees the footage, never where the turtles really are, and turns motion into pins on a map of the beach. Pins from both cameras are combined, so a site seen by both is more trustworthy.

The dashboard is built with Next.js and Postgres in Docker. Phones stream live video straight to the dashboard over Wi-Fi with WebRTC, and the motion view highlights movement right in the browser.

Challenges we ran into

Darkness. Turtles nest at night, and our very-dark mode (visible-light noise, not infrared) currently places no pins. Near-infrared is the next step.
Parallax. A turtle's body sits above the sand, so projecting it onto the ground shifts its apparent position. Two cameras facing each other help, but don't remove it.
Phones as cameras. Browsers only open the camera on HTTPS pages, so we added a local HTTPS proxy. We also had to handle WebRTC connection messages that arrive out of order.
Team integration. Partway through, the dashboard was rebuilt from scratch, and features had to move into the new app.

Accomplishments that we're proud of

In daylight, the two-camera pins found nearly every simulated dig site with very few false pins. We had set a perfect score as our target in advance and fell just short, and we report it that way instead of moving t
A detector that never sees the simulator's answers, so its results mean something.
An end-to-end system: from simulated sand to a pin on a live.
Turning any phone into a live beach camera with nothing to install.

What we learned

Measuring honestly matters more than a good headline number: a target you set in advance and miss teaches you more than one you quietly adjust. We also learned how much careful work it takes to separate what a syst*, and that a simple tool (a pin that says "look here") canchange how a search team spends its morning.

What's next for HatchWatch

Near-infrared cameras for real night conditions, field trials with local nest patrols, and solar-powered pole housings (the printable design is already in the repo).