Dots and sites
Found by colour and by shape: minimum and maximum radius, a roundness threshold. Optional character recognition reads the label sitting beside them.
Geomatics module · live
Your map is a picture.
Make it geometry.
Drop in a coverage map, a network plan or a survey — PDF or image. You georeference it, detection pulls out the objects by their colour, you correct whatever it missed, then you export in your GIS's own format.
Drag your map here, or click to choose it
PDF, PNG, JPEG, TIFF — 50 MB maximum
No file to hand?
File deleted 2 h after your last action
How it goes
4 steps- Step 1 Georeferencing. You click a point on the map, then the same point on an OpenStreetMap background. From three points on, GDAL computes the transform and shows the mean error in metres, along with each point's own deviation — the one that sits wrong is visible at once.
- Step 2 Detection by colour. One layer = one colour sampled from the map, and one kind of object: point, polygon or route. You add as many layers as the map has families of objects; layers already detected are kept.
- Step 3 Correction. Vertex handles to rework an outline, freehand drawing for whatever the colour missed, an eye to hide a layer and see the map underneath. Every object added by hand is projected through the same transform as the detected ones.
- Step 4 Export. KML, KMZ, GeoJSON, GeoPackage, Shapefile or GeoTIFF, with the accuracy report (method and RMSE) in front of you as you choose.
What detection can do
Point
Polygon
Coverage areas
Simplified outline, adjustable minimum area, halftone holes closed up. A lake left out of the coverage stays a real hole in the exported geometry, not only on screen.
Route
Fibre and linear networks
Reduced to their centreline, not their outline. A dashed line is joined back up within the gap you set, and a branching network comes back as one route per span.
Frequently asked questions
How do I convert a PDF map into a GIS file?
In four moves: you drop the map in, you georeference it by pointing at three or four landmarks on an OpenStreetMap background, detection pulls out the objects by their colour, you correct whatever it missed, then you export in your GIS's own format.
How many control points does georeferencing need?
Three at least: three is where GDAL can compute the transform. Beyond that, each extra point tightens the result and shows its own deviation, so a badly placed landmark gives itself away at a glance.
Which formats can the map be exported to?
KML, KMZ, GeoJSON, GeoPackage, Shapefile and GeoTIFF. The first five carry the objects that were detected and corrected; the GeoTIFF carries the map itself, warped and set on its coordinates, ready to serve as a backdrop in QGIS or ArcGIS.
How accurate is the georeferencing?
It is measured, not promised: the workshop shows the root mean square error in metres, along with the method used, and the accuracy report stays in front of you as you pick an export format. A badly fitted map says so before it is exported.
What happens when detection misses an object?
You draw it. Vertex handles to rework a rough outline, freehand for whatever the colour did not see, an eye to hide a layer and see the map underneath. An object added by hand goes through the same transform as the rest.
Do I need to install QGIS, ArcGIS or a plugin?
No. Everything happens in the browser, with no account and no install, and it is our server that cuts the background map into tiles: no other domain learns which area you are looking at. You then open the export in the software you already use.
Your map does not leave Frankfurt
The file is converted, analysed and exported on our own server. No third-party artificial intelligence service is called, and nothing trains a model. The working folder is destroyed two hours after your last action, even if you forget the tab is open.