3D Wireframe Generator from Blueprint Images

The project has reached a point where it makes sense to freeze the current version, prepare an export build, and leave it alone for a while.

The original goal was simple: verify whether several orthographic blueprint views could be turned into a usable 3D wireframe by manually matching corresponding points. That worked, and the workflow gradually became much more practical around that basic idea.

The program can now load a top view, side views, and optionally front or rear views. Reconstruction remains intentionally human-guided. The user decides which points in different views actually correspond to each other, while the program handles axis alignment, projection, and calculation of the resulting X/Y/Z coordinates.

A major improvement is the blueprint preparation step. The source PNG or JPG no longer has to remain just a passive image underneath the points. The program can remove a light background, reduce the drawing to thin centerlines, and split the result into selectable line segments. Unwanted parts can then be suppressed manually, so things such as mirrors, handles, wipers, dimensions, or other distracting details can be removed from the working drawing.

This creates a much cleaner reconstructed blueprint on which points can be placed more accurately than on the original scanned image.

The reconstruction workflow is also no longer limited to matching individual points. A section panel allows work with horizontal layers. A point can be selected in the side view, projected into the top view, and the user only needs to complete the missing coordinate. This makes it possible to build body profiles progressively at different heights.

Symmetry tools were also added. A manually created half or quarter of the model can be mirrored across longitudinal or transverse symmetry planes. Points that lie directly on the symmetry plane remain shared as the common seam between both sides.

Reconstructed points can be connected into real 3D edges. A separate 3D view displays points, edges, section planes, symmetry planes, and an experimental surface triangulation preview. The goal is still not a fully automatic black-box reconstruction. The user remains in control of which geometry is actually created.

After entering one known real-world dimension, the complete model can be recalculated to the required scale. Confirmed 3D edges can then be exported as ASCII 3D DXF for further work in a CAD or 3D modeling application.

The current workflow is roughly:

PNG/JPG blueprint → drawing cleanup → removal of unwanted lines → matching points or working with sections → symmetry → 3D points and edges → scale calibration → DXF export.

Several possible directions for further development have appeared during testing. One of them is matching entire cleaned lines between different views instead of manually matching individual points. The program could then generate points automatically along corresponding curves. It is not yet clear whether this would be more practical in real modeling work than the current combination of points, sections, and symmetry.

For now, the current version will be released as an experimental build.

The main question behind the project has already been answered: this reconstruction method works and can turn an ordinary blueprint into a useful starting point for a 3D model.

Further development is now less about whether reconstruction is possible, and more about finding which way of entering geometry is the fastest and least annoying in actual use.