A deep-water container terminal at Cai Mep in southern Vietnam, read from stock drone footage. Every frame has a solved camera position, so anything moving differently from the ground is a vehicle: thirteen trucks tracked in thirty seconds, each placed on the ground in metres. Our own demonstration, and the part of it that did not work is on this page too.
A terminal’s own systems know what was booked, gated and lifted. What they do not hold is the yard as a place: where the traffic actually goes, how full the blocks are getting, and what the site looks like from the angles nobody has a camera on. Those questions get answered by walking out and looking.
The commercial questions about a terminal are about space and movement: how much of the yard is working, what the traffic is doing in it, and whether the shape of the operation matches the plan it was designed around. Those are spatial questions, and the systems that run a terminal are not spatial.
How full the blocks are decides what can be accepted and where it can go. An independent read of it, even a rough one with its range on display, is a different kind of input from the same number out of the same system.
Where vehicles actually drive and how fast tells you how the layout performs. It is the kind of thing a site plan cannot answer and a shift supervisor cannot be expected to quantify.
Fixed cameras see what they were pointed at years ago. A flight sees the land side, straight down and the berth in one pass.
A terminal is a shift pattern, not a moment, which is why anything read from a single flight is a snapshot and has to say so.
Thirteen vehicles tracked in thirty seconds, each with its path on the ground in metres and its speed.
Counted by AI on 338 views of the real frames and paired with the yard area measured on the 3D.
The land side, straight above the yard and the evening, each as its own 3D rather than a forced merge.
Fly it twice and the second pass answers moves per hour and dwell time, which one pass cannot.
With the honest caveat that one flight is a snapshot rather than a shift, and every figure below carries its range.
Every figure on this page says what kind of figure it is. That is the whole of why it can be used.
From this footage. That is stated here rather than worked around, and it is the reason there is no demo to open.
The box count is our estimate as a range. Their count is theirs and we do not have it.
The yard area everything else rests on was measured on a 3D carrying about ten per cent. A surveyed area beats it, and so does the terminal’s own record of its own yard.
One flight measures what moved while it was flying. Anything about rates over a day needs repeat flights.
Nothing about time or cost in a real terminal was measured here, so there is no figure for it.
Built on stock footage, as our own demonstration. Nobody commissioned it: it exists to show what the method does on a working industrial site rather than on a tidy one. The site and its operator are not named here and are not ours to name.
The 3D. The stock clips are long, straight, high passes, so the stacks and the cranes never get enough angles and the reconstruction came out too soft to put in front of anyone. The tracking and the counts stand, because both are read from the real frames rather than from the model. The 3D needs a proper orbit flight of the yard, and until that happens there is no demo of it.
Because the alternative is a page that implies all of it worked. Which half of a method holds up on which kind of footage is the single most useful thing we can tell somebody thinking about flying their own site.
No, and it must not be read as one. It is an estimate from an AI count of the stacks and the measured yard area, published as a range of 7,600 to 12,900 TEU. The terminal’s own figure is the terminal’s, and we do not have it.
They are estimates. Every frame has a solved camera position, so the software knows how the ground should move and anything moving differently is a vehicle. Turning that into metres per second relies on the 3D’s scale, which carries about ten per cent.
It is enough to measure what moved in those thirty seconds and not enough to describe a shift. Moves per hour and dwell time need repeat flights, and the page does not pretend one pass gives you them.
Stack heights read off the 3D were too noisy to count boxes, so the counting moved onto the real frames. Water basins were first picked up as moving vehicles, because still water moves differently from the ground; only solid ground is searched now. And the first 3D only saw the yard from the berth side and looked one-sided.
Stock footage of a working deep-water terminal. The terminal, its operator and the shipping lines are not named and are not ours to name. Nobody commissioned this.
Orbits at two heights over the yard rather than straight passes, which is the whole of what the 3D needed. Then the reconstruction is worth publishing and the counts get a model to sit on.
Tell us what it is and what you need to know about it. We will say whether this would answer it, including when it would not, and this page is what that sounds like.