DRONE PHOTOGRAMMETRYComing soon
Stream Capture is photogrammetry software for Windows that turns overlapping drone photos into an orthomosaic, a DSM and DTM, a classified point cloud and a textured 3D model, then tells you how accurate they are. It runs on your own GPU, works offline, and keeps check points apart from control points so the accuracy it reports is independent evidence.
Stream Capture is in development and not released yet. At launch it will come in the same download as its sister program Stream Terrain; today's download contains Stream Terrain.
Photogrammetry measures from photographs. Overlapping photos are matched to one another, the position and lens of every camera are solved, and the geometry of the scene is recovered by triangulation. From that comes a point cloud, a surface, a map-accurate orthophoto and a 3D model — everything a drone survey is delivered as.
Stream Capture runs the whole chain as seven stages — Data, Align, Dense Cloud, Mesh + Texture, DSM, DTM, Orthophoto — and writes the products, a quality report and the evidence behind it to a project folder you choose. Source photos are never modified.
Does it report error at independent check points? Fit at the control points it was forced through says little. Look for a separate check-point error, and for an accuracy class stated from check points only.
Does it label what it filled? A surface model with no gaps is suspicious. Look for a mask that says which cells were measured and which were interpolated, and a way to see how far to trust each cell.
Does it give the same answer twice? Reproducible output — the same photos and settings giving the same files — is what lets you trust a change between two surveys.
Where does your data go? Some tools upload photos for processing. If your photos are confidential, or the site has no connection, check that processing and the licence work offline.
Which GPUs does it need? Some photogrammetry is tied to one graphics vendor. A vendor-neutral GPU path avoids a hardware decision you may regret, and a CPU fallback keeps a broken driver from stopping work.
Do the outputs share a grid? An orthophoto, DSM and DTM written on one cell lattice overlay in GIS without resampling and without a half-pixel shift.
Can it queue, pause and resume? Long jobs fail for ordinary reasons — a closed laptop, a restart. Look for finished stages that are kept and a queue that survives a reboot.
What does it say it does not do? A tool that lists its limits is usually one you can believe about the rest.
Add the photo folders; the project scan reads the metadata, counts RTK-fixed frames, flags blurred frames and missing positions, and draws a coverage map. Confirm the coordinate system, add ground control points if you have them, choose the products and one of three presets — Fast Draft for field checks, Balanced for most work, Maximum for full-resolution delivery — and process.
The orthophoto, DSM and DTM are GeoTIFFs on the same grid. The point cloud is LAZ with classes, with COPC, CSV and 3D Tiles available. The 3D model is OBJ or GLB with a best-view texture atlas. Contours export as DXF, GeoJSON, KML, Shapefile or GeoPackage, and the terrain as a LandXML surface.
The quality report shows GCP RMS and check RMS side by side, the reprojection error at full resolution, the height error of the DSM and DTM at the points, the camera calibration and an ASPRS accuracy class from the independent check points. When there are fewer than 30 check points it says the class was tested with too few.
Without ground control, a systematic offset between the drone's frame and your ground survey — usually in height — can remain. With control points applied, independent check points measure what is left, and the report gives horizontal and vertical RMS from those points only.
Stream Capture is a native Windows application. It does its heavy work through Vulkan on NVIDIA and AMD GPUs, with a CPU fallback that uses the same methods, and it works out of core so large projects stay inside your RAM budget.
There is no upload step and no per-project quota. Queued projects resume from the stage where they stopped after a restart, and the interface and the PDF report come in six languages.
It processes photos from RGB frame cameras in JPEG, TIFF and PNG. It does not process multispectral or thermal radiometry, fisheye or spherical lens models, or video, and it has no Mac or Linux version. It does not compute PPK corrections from raw satellite logs; it uses the corrected positions already in your photos. It is not a cloud service.
Stream Terrain is the analysis side: it streams very large DEMs, orthophotos and point clouds and computes volumes, cross-sections, stockpiles, slope, hydrology and change detection. Stream Capture writes exactly the files Terrain opens, and Send to Stream hands a finished project over in one step.
Every product and every proof has its own page, with the formats it writes and the questions people actually ask.
The whole chain from photo folder to deliverables, with its health check, queue and presets.
True-ortho mosaic with ghost-free blending, a source map and GeoTIFF, PNG, KMZ and MBTiles output.
Elevation models with honesty masks, confidence maps and LandXML export.
A dense, classified point cloud on the GPU, written as LAZ, COPC, CSV or 3D Tiles.
A textured 3D model with a best-view atlas, as OBJ or GLB.
Automatic target marking, a GCP editor and check points kept apart.
RTK and PPK positions from the photo metadata, and the accuracy you really get.
Check RMS and an accuracy class from independent points.
Vulkan on NVIDIA and AMD, with a CPU fallback and reproducible output.
Open and analyse the elevation model in Stream Terrain.
Stream large orthophotos as a basemap in Stream Terrain.
Open and rasterise the LAZ point cloud in Stream Terrain.
Volumes between two surfaces, on the surface you just produced.
Stockpile volume and tonnage from the drone surface.
Compare two flights of the same site.
Mapping with a small consumer drone: soft GPS priors, rolling shutter and the role of ground control.
Software that measures from photographs: it matches overlapping photos, solves where each camera was, and recovers the 3D shape of the scene to produce point clouds, elevation models, orthophotos and 3D models. Drone mapping is its most common use.
The one that proves its accuracy on your kind of project. Use the eight questions above: independent check points, labelled gaps, reproducible output, where your data goes, which GPUs it needs, shared output grids, queue and resume, and a stated list of limits. Stream Capture is designed to answer each of them in the open.
A GPU is not required but it makes large projects practical. Stream Capture uses the GPU through Vulkan on NVIDIA and AMD cards and falls back to the CPU with the same methods, only slower.
Not at present. Stream Capture is a Windows desktop application.
Yes. Processing runs on your computer, nothing is uploaded, and the licence works offline.
Photogrammetry gives dense colour and a precise surface wherever the camera sees it and needs only a camera. LiDAR can see through gaps in vegetation and measures its own light. For open ground and structures photogrammetry is usually enough.
Stream Capture is in development and not released yet. At launch it will come in the same download as its sister program Stream Terrain; today's download contains Stream Terrain.
Get notified at launch