Georeferencing CCTV and Drone Video: What Changes?

Video georeferencing is the work of converting points seen on a CCTV or drone screen into real map coordinates. If you know where the camera is, which way it is looking and its field of view, you can calculate where a point on the screen falls on the ground.
In short
- Video georeferencing converts points on a screen into map coordinates
- If you know the camera position, direction and field of view, you can calculate ground coordinates
- Terrain, camera shake and input errors reduce accuracy
Suppose you are looking at a CCTV screen showing a wooded hill where a wildfire has broken out. Smoke rises from the upper right of the screen, but the person in the control room cannot immediately tell exactly which village's back hill it is, or whether a fire engine can reach it within a few minutes. A single pixel on the screen is not a coordinate on a map; it is just "upper right."
Video georeferencing fills this gap. If you know where the camera is and which way it is facing, you can convert a point on the screen into real latitude and longitude. In a disaster control room watching dozens of CCTV and drone feeds at once, this one conversion changes "what do I see on which screen?" into "what is happening where on the map?"

What exactly is video georeferencing?
Georeferencing is the work of giving a map or an image a coordinate system. A typical example is attaching latitude and longitude values to a satellite image. Video georeferencing goes one step further and converts each point on the screen of a video taken at an oblique angle, such as from CCTV or a drone, into map coordinates. A satellite image usually looks straight down, so attaching coordinates is relatively simple, but a ground CCTV looks at an angle, so distances on the screen and distances on the ground are not constant. At the bottom of the screen (close to the camera), a few meters take up a certain number of pixels, but at the top (far away) the same number of pixels may cover tens of meters. So you cannot simply convert by the screen ratio, and you must also calculate the geometry of the camera. With this process, the control room can point at "the spot where that smoke is seen" directly by address or coordinates.
If you know the camera position and direction, do you get the coordinates?
Three kinds of values are needed for the calculation. The first is the camera's installation coordinates (latitude, longitude and altitude). The second is the direction the camera faces (azimuth) and the angle it tilts downward (tilt). The third is the field of view (the width of the view the lens captures). If you add these values together with information on ground height (a digital elevation model), you can work backward from a pixel on the screen, through the camera, to the point where it meets the ground. Put simply, it is like knowing the direction and angle you hold a flashlight and so knowing where on the floor the light falls. For drone video, GPS and attitude sensors (gimbal angles) are often recorded in the video file, so this calculation is relatively easy, but for fixed CCTV these values must be surveyed and recorded when it is installed in order to calculate automatically later. That is why it matters to keep this metadata from the start when installing new CCTV for monitoring landslides or river flooding.
Why is it good to gather many videos on one map?
A disaster control room usually receives dozens to hundreds of CCTV feeds, plus drone video, at the same time. Opening the screens one by one makes it hard to grasp "where is dangerous right now" at a glance. If you apply georeferencing to each video, you can overlay the observation range of every camera and the events currently visible (fire smoke, flooded sections, landslide debris and so on) as points or polygons on a single map. This also allows cross-checking whether the same event is confirmed from both sides where camera views overlap, and conversely it immediately shows where the blind spots are. When deciding evacuation routes too, you can judge while checking on the map directly that "the road this CCTV shows is already flooded," which speeds up decisions compared with receiving the situation one message at a time over a communication line.
Why does the position not match exactly and drift?
A georeferencing result always carries some error. The biggest factor is terrain. If the digital elevation model used in the calculation differs from the actual ground height (especially in areas with many buildings or trees), a point on the screen is calculated as meeting the ground at the wrong altitude, and the position drifts. The second is camera shake or zoom changes. If the CCTV sways in the wind or the operator zooms in, the field of view and direction values recorded at installation no longer match. The third is the drone's GPS error or attitude sensor error, where momentary signal instability during flight can lead to coordinate errors. The fourth is that the farther the distance, the more a small angle error grows into a much larger distance error on the ground. So in situations that need a precise boundary, such as a flooded section, instead of trusting the coordinates calculated automatically from the video as they are, you need a procedure to check them against field coordinates or other video.
How do you check that the video you are viewing is properly aligned?
The simplest way to check yourself is to pick one reference point whose location you already know on the screen (an intersection, a building corner, a bridge and so on) and compare by eye whether that point matches the coordinates shown on the map. If only one reference point matches and other points are off, the field of view or tilt value is probably wrong, and if everything is shifted in one direction, the camera coordinates themselves are likely wrong. When connecting a new CCTV to the map in a control room, it is safer to cross-check with two or three or more reference points like this before using it for actual response decisions.
Step by step on screen
The image is a diagram of the order of steps. The original does not name the GIS program, so menu names may differ from those described.

- Check reference point coordinates
On a public map (such as V-World or Kakao Map in Korea), write down in advance the coordinates of an intersection or building near the CCTV. - Enter the camera information
In the GIS program, add the camera position point with Layer > Add Layer > Add Vector Layer. - Draw the field of view
In editing mode, use the field of view and direction values to draw the triangle the camera covers as a polygon. - Compare with the video screen
Put the reference point on the CCTV screen and the coordinates shown on the map side by side and check that they match. - Record the error
In Open Attribute Table (right-click the layer > Open Attribute Table), note the error distance for each reference point. - If menu names differ
Menu names and positions may differ by version, so check the help or the right-click menu in the Layers panel.
Menu names may differ slightly between versions. The figure above is a diagram that summarizes the order of steps.
Prompts you can copy
When you want a step-by-step checklist for how to start georeferencing CCTV or drone video
Your role is a GIS technical advisor helping a disaster control room geospatial officer. I am about to start connecting [CCTV or drone] video to map coordinates for the first time. The information I currently have is [camera installation coordinates, whether I have the azimuth, whether I have a digital elevation model]. Answer in the following format. 1) A list of values I need to additionally survey or confirm 2) Realistic ways to confirm each value (including ways that need no surveying equipment) 3) Points where error may grow, and why 4) Criteria for choosing reference points to check when verifying the result When you use a technical term, explain it simply in one line. Do not guess numbers or product names you are not sure of; answer in general terms.
Replace the bracketed parts for your situation. Do not use an answer as it is; check the facts first.
Try it yourself
- On a public map service, find the locations of CCTV near your home (public information such as the Public Data Portal) and estimate which way they face
- Take a photo out of your window with a smartphone camera and compare by hand a building in the picture with the location of that building on the real map
- If you have used a drone, check in the app whether GPS and gimbal angle values are recorded together in the flight log
Common questions
Is georeferencing easier for drone video than for CCTV?
Generally yes. Drones often record GPS and attitude sensor values automatically together with the video, so there is less need to survey the values needed for calculation separately. However, signal instability during flight can introduce error into those values themselves, so checking is still necessary.
Can you not do video georeferencing without a digital elevation model?
In areas close to flat ground, a simple flat-ground assumption can give an approximation. But in mountainous areas or areas with many buildings, height differences are large and a flat-ground assumption alone can produce big errors, so it is safer to use terrain information if possible.
Can georeferencing be applied later to CCTV that is already installed?
Yes. If the coordinates or direction values were not recorded at installation, you either go to the site and survey again, or compare several reference points on the screen with the map and estimate the camera conditions backward.
What comes next
Video georeferencing turns scenes on a screen into facts on a map, so it is especially powerful when a control room handles many videos at once. A later post in this series covers how AI can automatically find events such as fires and flooding in video that has been given coordinates in this way.
This is an English adaptation of a post in the Korean GeoAI basics series. The original is here, and more learning materials are on GeoAI for Disaster Safety.
This post was drafted with the help of AI tools and reviewed by the author. Tool features and data portals change over time, so check the official documentation when you practice.
Comments
Post a Comment