3D Reconstruction Using Active Illumination

The active illumination sphere.

Using active illumination for 3D object reconstruction is an established technique. It is widely used in quality control, reverse engineering, object recognition, medical applications, historical artifact conservation, animation, and virtual reality. One of the currently most promising approaches for high-resolution, high-speed, low-cost, non-contact, and non-destructive 3D surface reconstruction is a combination of structured-light scanning with photometric stereo. Our previous work has shown that this technique makes it possible, for example, to scan a human face in a few seconds at a resolution such that individual pores and fine wrinkles of the human skin can be recovered.

In our previous work a combination of structured-light scanning with photometric stereo was used for facial appearance capture: sequence of illumination patterns (top), resulting high-resolution 3D normal field reconstruction (bottom).

However, state-of-the-art approaches have problems with occlusions, which make the scanning of objects with complex topology difficult. In contrast to existing work, our project’s objective is the development of algorithms that are robust to occlusions. This will enable high-resolution scanning of objects with complex topologies, such as green plants or man-made objects with disconnected components, and will lead to new applications (e.g., automatic monitoring of green plant growth, quality control in industrial production, etc.).

Arrangement of the RGB-LEDs inside the sphere.

For the generation of illumination patterns, a large spherical metal frame (diameter 216 cm) is constructed from two moveable geodesic domes. The sphere is equipped with a total of 2202 RGB LEDs. Using a microcontroller each of these LEDs can be individually regulated and each can emit a mixture of red, green, and blue light. Driving all LEDs of the sphere requires only 6.2 milliseconds (160 Hz). Therefore, it is possible to display various lighting patterns in a very short time, which enables measuring dynamic objects as well.

 

Project team

Principal Investigator
Prof. Dr. Thorsten Thormählen

Researcher
Martin Grochulla
Christian Kurz