At about 1:28 in the morning on March 26, 2024, a camera mounted on a friend’s waterfront home recorded the container ship Dali losing its lights, regaining them, losing them again, and then hitting the Francis Scott Key Bridge. It was the work of a hobbyist, Mike Singer of Anne Arundel County, who had put it up for his Baltimore and Chesapeake Bay Ship Watchers group. By breakfast the clip was on every network, and the Maryland delegate who later honoured Singer called the footage instrumental to the investigation.
That is the case for port webcams in one paragraph. They are cheap, they are always on, and they sit on exactly the stretches of water that matter: harbour mouths, straits, naval piers, anchorages. AIS tells you what a ship says about itself. Satellites give you a frame every few hours or days if you’re lucky with cloud. A webcam gives you continuous video of a fixed piece of water, and nobody aboard the ship gets a vote on whether it is switched on.
Governments know this, which is the other half of the story. In 2017 the Swedish Navy asked Karlskrona, home of the country’s main naval base, to move a tourist camera on a water tower that looked straight over the harbour and its approaches; the council turned it toward the islands of Dragsö and Långö. In 2023 the Navy had the National Park Service pull two privately run cameras from Cabrillo National Monument in San Diego, a few months after one of them caught the destroyer Momsen and the dock landing ship Harpers Ferry in a near-collision that the internet named “warship chicken”. NCIS cited force protection concerns covering Naval Air Station North Island, the submarine piers at Point Loma and Naval Base Coronado. In January 2024 Ukraine’s SBU said Russian services had hijacked two residential cameras in Kyiv, re-aimed one and pushed its stream to YouTube to watch air defences during the January 2 missile and drone attack. The SBU says it has blocked about 10,000 cameras since February 2022.
Finding the cameras and pinning them down
Start with the people who have a reason to point a camera at the water. Port authorities run cameras for pilots and the public; so do harbour masters, marinas, yacht clubs, coastguard stations and ferry operators. Municipal tourism boards put cameras on towers and hotels because a harbour view sells rooms. Then there are aggregators, which collect thousands of public streams by location, and YouTube Live, where ship-watching channels run 24-hour feeds of the Elbe, the Bosphorus, Rotterdam’s Nieuwe Waterweg, the Solent and the Panama Canal locks. Some vessel-tracking sites embed nearby cameras next to their AIS maps. Search in the local language: hafen webcam, havnekamera, liman canlı, and add the name of the specific quay or lighthouse rather than the city.
The Bosphorus is the model case. Istanbul’s ship spotters, analysts like Yoruk Isik among them, have for years combined their own photographs with coastal webcams, smartphone AIS apps and social media posts that show a grey hull in the background. Warships often switch their transponders off through the strait; a camera on the shore doesn’t care. Our sister site bosporus.org covers the strait itself.
Once you have a feed, fix three things before you trust anything it shows. Location first: match fixed features (breakwaters, cranes, a church spire, light structures) against satellite imagery and nautical charts until you can put the camera on a specific roof. Then direction and field of view: draw the cone on a map, with two landmarks at the left and right edges of the frame, so you know which berths and which part of the channel are actually visible. Then scale: a known ship length or a known berth length lets you estimate the size of anything else that passes. Write these down once per camera; most feeds don’t move.
Timing, correlation and the dark arrival
A webcam frame is only evidence if you know when it was taken. The overlay clock on a stream is only a claim, so check it against something independent. A ferry with a published timetable is the easiest anchor. AIS works too: if a ship broadcasting its position passes the camera’s field of view at 14:07 UTC by its track, the frame showing that ship is 14:07 UTC, whatever the overlay says. Shadows give you a cross-check for daytime footage with sun position tools. Record everything in UTC and note the camera’s local time zone and daylight saving state separately, because a lot of timing errors come from a stream that says 09:00 without saying where.
Correlation is where webcams earn their keep. Lay the camera’s visible cone over an AIS map for the same window. Every hull that crosses the frame should have a matching track; a ship on camera with no corresponding AIS position is a dark arrival, or a vessel spoofing somewhere else, or a vessel too small to carry a Class A transponder. Size and silhouette usually settle which. The method is the same one used to match Sentinel-1 radar detections against AIS, which we covered in our guide to AIS, satellite imagery and port activity, except with a sensor that refreshes every second instead of every few days. Satellite imagery then confirms or extends what the camera saw: where the ship berthed, how long it stayed, whether the camera’s anchorage count matches the one from orbit. For chokepoint work in the Gulf, the same layering runs through our coverage on hormuz.net; for commercial traffic patterns, see maritimeshipping.org.
Archive as you go, since live streams vanish. Record the stream with a tool such as yt-dlp or a screen recorder, keep the original file untouched, and log the page URL, the stream ID, the capture start time in UTC, and the software you used. Hash the file (SHA-256 is fine) at capture and record the hash in your notes; work only on copies. If a clip later matters, you can show it has not changed since the day you grabbed it.
Where webcam evidence breaks
The common failures are dull and frequent. Stale feeds keep serving the last good frame after the camera drops off the network, sometimes with a ticking overlay clock. Looped footage is worse: an operator, or someone with access, replays an hour of empty harbour on repeat. Look for the same car crossing the quay at the same second each hour; compare the light against the actual sun position for the claimed time; check whether the weather on screen matches the weather reported for the port. Compression destroys the detail you most want. A pennant number or a ship’s name at 600 metres becomes four grey pixels on a 720p stream, and heavy encoding can invent edges that look like antennas. Don’t read hull numbers off a compressed frame unless you can confirm them another way.
Then the physical limits. Resolution falls off fast with range, so most cameras are good for type and silhouette, not identity, beyond a kilometre or so. Fog, rain on the lens and sea haze kill the image just when ships are moving under poor visibility. Night reduces most feeds to navigation lights, which still tell you course and rough size, and not much else. And in a crisis the cameras go away. Karlskrona, San Diego and Kyiv all show the same reflex: when a camera starts showing something sensitive, it gets turned away or switched off. A camera that disappears is itself a data point; note the date.
The legal and ethical lines are real and they vary. Watching a public stream is legal almost everywhere. Republishing it may not be: in Ukraine, publishing footage of defence force activity can carry up to 12 years in prison, and other states at war or under threat have similar rules. Accessing a camera that wasn’t meant to be public (default passwords, exposed admin panels) is unauthorised access in nearly every jurisdiction, whatever the intent. Stay with what the owner chose to publish. And think before you post a timestamped clip of a warship leaving port during an active conflict; the same frame that verifies a claim for you can serve as a targeting cue for someone else. The camera on the Bosphorus is a public resource. What you do with it is your responsibility.