VaxRecon3D technology was developed by Vaxtor Systems for the Avatar 3D project. This is a TELEFONICA Technology Innovation Project software capable of reconstructing the tridimensional model of a person or object from video recorded on a mobile phone or conventional video camera.
Current 3D reconstruction technologies consist of projecting different types of light beam on the object. Depending on the type of light (laser, infra-red, modulated, structured...), different types of information are obtained such as the distance to a point or deformation of a light pattern, which are used as the basis for calculating the reconstructed model. It is also possible to carry out reconstructions using a stereoscopic system or by using multiple cameras.
However, VaxRecon3D is a completely non-invasive technology that uses a single video camera. The 3D reconstruction is based on the movement of the actual camera; the objects present in the scene are modeled based on a video sequence.
As with the other technologies, VaxRecon3D requires a controlled environment to operate properly: static background, video recording based on certain conditions, etc.; small rules that may be implemented and followed easily and from which no current system used for reconstruction may escape. The advantage of VaxRecon3D is that only a single camera is required, a cell phone camera is sufficient; the task is thus simpler and may be carried out in any location.
At the present time, non-invasive (video only) 3D reconstruction technology is an open research niche that, for the moment, is unable to provide a definitive, reliable technique that provides completely satisfactory results. VaxRecon3D represents the base technology on which our future efforts will be based in order to meet our objective of providing faster, more reliable and realistic 3D reconstructions in less controlled conditions.
The videos that are currently being used for 3D reconstruction are between 30 and 60 seconds long. The scene must be static, i.e. objects in the scene must not move. A high resolution video camera is not necessary for video recording. Indeed, all models shown above has been built from video recorded with an iPhone.
The time required for reconstruction depends on scene complexity and the required point cloud density (usually between 20,000 and 500,000); however, we are able to state that the average time required for reconstruction is less than 15 minutes. Given the nature of the methods and algorithms used by our system, it is possible to reduce these times substantially and this is one of our current lines of work
Reconstruction examples using VaxRecon3D and iPhone video
The original video scene can be seen on the left, images of the reconstruction on the right. Some surfaces have been meshed; others show points cloud reconstruction only.