Pixel Scanner

PixelScanner is our solution to automated 3D modeling from images. It is the end product of many years of research and one of the most efficient solutions of its kind. It processes hundreds of images and outputs camera calibration, a dense 3D point cloud and a high quality textured mesh of a scene in minutes, with no user intervention.

Four stages

1. Image capture. The only step that involves the user.
2. Camera calibration. Geometric and photometric information about the data set is extracted.
3. Point cloud computation. A dense point cloud of the object is extracted using the calibration.
4. Mesh model extraction. A triangulated mesh is computed and textured with the captured images.

Camera calibration

Calibration finds the true parameters of the camera that produced each image: internal parameters such as focal length, principal point and lens distortion, and external parameters that encode the position and orientation of the device in the world. Using image information alone, PixelScanner finds both and outputs a sparse model of the scene. Calibration can be extended to metric scale for measurements on the final model, and if GPS data is available, geo-referenced calibration is generated automatically.

Point cloud computation

A point cloud is a set of vertices in 3D space representing an object, usually produced by laser scanners that bounce a beam off a surface. PixelScanner emulates this using only captured images, generating a very dense oriented point cloud without any active device. Any textured object can be modeled from photos, even from far away, such as from a helicopter or drone. That makes it possible to model objects in settings where a laser scanner is impractical or too expensive.

Besides colored dense clouds, it extracts sampling density, curvature and surface normals at each vertex, so models can be used directly in graphics, industrial metrology or digital elevation models. Users can set the sampling density or point count, and clouds can be down- or up-sampled without losing surface structure.

Mesh model extraction

Triangle meshes are the most common 3D representation in games, animation and any application that uses 3D models, and today's computers include dedicated hardware to render them. PixelScanner automatically generates high quality textured meshes from the extracted point cloud. Mesh and texture resolution can be controlled independently to fit any rendering budget.