home
Extract 3D point cloud from photographs
This page provides support to reconstruct a 3D model from a series of 2D photos using Photoscan, VisualSFM and COLMAP.Upload your own images for review here.
Below an example of a 3D model of a carton box extracted from a set of photos using VisualSFM.
The following steps are needed to extract a 3D object from a set of 2D images:
- Collect 2D image dataset for static (not moving) object
- Feature detection and extraction from (separate) images
- Distinctive feature description (2D key points)
- Match features between images (3D tie points)
- 3D model reconstruction
Align Photos in COLMAP
To get started in COLMAP (windows version): Start a new project, enter a name for the new database, and select the folder for your photos.Feature Extraction
Click 'feature extraction' in the Processing menu to open the Feature Extraction options and settings.
There are quite some options here but normally you only need to adjust the max_image_size. Reduce the max_image_image_size if the default setting results in too high memory use or too long processing time.
Increase the max_image_image_size if the default setting results in too few features being extracted.
You start feature extraction by clicking Extract.
The result (number of features) can be checked by clicking 'database management' in the Processing menu.
Feaure Matching
Click 'feature matching' in the Processing menu to open the Feature Matching options and settings.
The resulting matches between images can be checked in the match matrix (menu Extras) and in the database (menu Processing).
Images that have no matches normally have not sufficient overlap (or too large view angle difference) with other images. The best solution for that is uploading additional images but if that is not possible you can try increasing the max_error or decreasing the confidence.
Align Photos in Agisoft Photoscan
To get started in Photoscan Standard (demo mode): Open the 'Workflow' menu, add your photos, and open the 'Align Photos' menu.Align Photos - General Settings
Start with accuracy low. If (some) cameras are 'not aligned' then that probably indicates lack of overlap between your photos.
Increasing the accuracy setting to medium or high will normally not result in more aligned photos.
Deselect 'generic preselection' (only switch this on if you want to reduce the processing time).
Select 'reset current alignment' if you re-do an alignment. Reset off does not seem to work in demo mode.
Align Photos - Advanced Settings
With accuracy set to low, deselect both key point and tie point limit settings by using zero value or by leaving the fields empty.
Enable adapative camera model fitting.
extract3d 2018