On this page

    How to capture drone data for GeoSplats

    This guide explains how to fly a drone and capture aerial video or photos suitable for GeoSplats (3D Gaussian splatting). It covers what to plan before you fly and what to do while you’re in the field. Once you have your footage, see how to prepare your splat model: that’s where you turn it into a 3D model, clean it up, and upload it.

    1. Select capture method

    You can capture drone data as timed interval photos or as continuous video. Generally, photos give you more detail, a lower risk of motion blur, and automatically embedded coordinates, while video is faster to prepare for and shoot, and provides denser overlap.

    Both methods can produce good, high-quality data for your 3D models when set up correctly, so the choice mostly comes down to your preferences.

    Timed interval photos

    Timed interval photos are a simpler, classic approach: the drone flies a pre-planned grid and takes a photo automatically every 1 to 2 seconds.

    It takes longer to set up and fly than continuous video, but you get sharper, higher-resolution images, and each photo saves its own GPS location and camera angle automatically.

    Continuous video

    Continuous video is the faster and more practical option. Instead of setting up an automated flight path, you fly smooth manual orbital rings or straight sweeps while recording.

    Modern drones (like the DJI Mini 4/5 Pro, Air 3, or Mavic 4) record their exact GPS position and camera angle throughout the flight and save that data into a companion .srt subtitle file alongside the video. To turn this on in the DJI Fly app:

    1. Open Settings (...)Camera tab.
    2. Under Advanced Settings, toggle Video Subtitles: ON.
    
    SD card directory
    ├── DJI_0001.MP4    <-- 4K+ video
    └── DJI_0001.SRT    <-- Timestamped GPS location and camera angle
    

    Later, when you feed the footage into the software you use to build your splat model, it reads a few frames per second from the video and matches each one with the GPS position and camera angle from the .srt file.

    2. Set up drone camera

    Set these before you take off, and don’t change them mid-flight:

    Parameter Recommended value Why it matters
    Zoom ratio Fixed / constant Changing the zoom partway through the flight throws off the 3D model.
    Shutter speed 1/500s to 1/1000s Keeps your shots sharp even while the drone is moving.
    ISO 100 (locked) Keeps dark areas clean, without noise.
    White balance Manual (5500K sunny / 6500K cloudy) Stops the colors from shifting as the drone turns.
    Focus Manual focus (MF) set to infinity (∞) Stops the camera from refocusing mid-flight and blurring your shots.
    Resolution 4K+ @ 60 fps or 20–48 MP photo Higher resolution feeds more visual detail into the reconstruction, giving you noticeably sharper textures and a better quality 3D output.

    If you don’t have time to set manual exposure, auto mode on 4K+ video works well too. Just keep two things fixed: the zoom level for the entire flight, and a slow flying speed (~3–5 m/s).

    3. Fly the capture pattern

    Which pattern to fly depends on what you’re capturing: fly tiered circles around a single building, or a grid pattern over an open area.

    Selecting and flying a pattern correctly ensures that the software can build an accurate 3D model. The software needs to see every part of the scene from several different angles: at least 3 to 4 for each feature. That’s how it figures out where exactly each surface sits in space, including tricky details like reflections. Otherwise you end up with blurry patches, gaps, or stray floating bits in your finished model.

    Remember that your 3D model can only show the detail you actually captured, so fly as close as you want your audience to be able to see.

    Single building

    For a building or other standalone structure, fly three circles around it at 3–5 m/s, keeping it centered, then finish with one diagonal pass that ties everything together.

    Orbital flight pattern for capturing a standalone building

    1. High orbit (~1.5x the building’s height, camera at -45° to -60°): captures the roof, chimneys, and the surrounding ground.
    2. Mid orbit (~0.8x the building’s height, camera at -25° to -35°): captures the walls, windows, and balconies.
    3. Low pass (2–5 m above the ground, camera at 0° to -15°): captures the base of the building, doors, and the underside of any roof overhangs.
    4. One diagonal pass (camera at -45°): a full circle at a steeper angle, to fill in anything the first three passes missed.

    When flying this pattern around intricate architecture, it’s usually more practical to navigate the drone by hand. You can flexibly adapt to tight spaces, keep the building nicely centered, and easily catch walls and roof overhangs that an automated top-down grid might miss.

    Open area

    For larger sites, like a campus, golf course, or park, fly a crosshatch pattern instead: two overlapping grid passes, one crossing over the other at a right angle.

    We recommend setting up an automated (programmed) flight path where you set the flight grid, height, camera angle, and overlap, and let the drone do the job. Make sure you set your flight altitude safely above tall trees, power lines, and hills.

    Crosshatch flight pattern for capturing open terrain

    Point the camera at an angle rather than straight down, around -45° to -60°, so you also capture tree trunks, wall textures, and slopes in the terrain that a straight-down shot would miss.

    Keep 75–80% overlap between passes, at a steady 3–5 m/s.

    Example: combining patterns

    This aerial survey of Helfštýn Castle combines both patterns in one flight: circling orbits around the central tower and ramparts, plus angled grid passes over the outer courtyards.

    Real-world drone flight survey over Helfštýn Castle

    The pilot flew a 19-minute continuous 4K capture covering the entire site. The resulting 3D model is available in the live GeoSplats showcase.

    Tips & tricks

    A couple of extra factors that affect how clean your capture turns out: the conditions you fly in, and how you handle surfaces that don’t reconstruct well in 3D.

    Light and weather conditions

    Overcast days are best for heading out with the drone. Mild, cloudy days give you soft, even light without distracting shadows. Sunny days work well too if you fly around midday when shadows are shortest. Avoid early morning and late afternoons when sun creates long, moving shadows that can confuse the 3D software.

    Fly on dry, relatively calm days. Strong wind makes smooth manual circles tricky, and water droplets on your camera lens will ruin your shots.

    Handling reflective surfaces

    To reconstruct a scene in 3D, the software looks for distinct visual details it can match across different shots. A few surfaces need extra care:

    • Water and fresh snow: Open water (lakes, rivers, oceans) and smooth, untouched snow don’t have many fixed visual details. While 3D Gaussian splats handle reflections much better than classic 3D meshes, if a photo shows only flat water or plain snow, the software won’t know how to connect the data. Make sure every shot also includes textured land features like shorelines, docks, trees, or rocks.
    • Glass and windows: Large glass facades reflect the sky and trees differently from every angle. Flying dense, multi-tier circles around the building gives the software enough different viewpoints to figure out what’s a reflection and what’s the actual structure.

    Flying legally

    Before you take off, make sure you’re familiar with the drone rules in the area where you’re flying:

    As a rule of thumb: steer clear of airports and crowds, respect private property and people’s privacy, and keep an eye on local no-fly zones.

    Next steps

    Once you’ve captured your footage, prepare your splat model: turn it into a .ply file, crop it, and upload it into your MapTiler account.

    Was this helpful?
    How to capture drone data for GeoSplats
    Drone capture