Range Camera Imaging : From Human Body Measurements to Very Large 3D Points Scenes Visualization
The use of camera range imaging for the purpose of 3-dimensional representation and non-contact measures is widely used in industry, medicine, computer-aided design and other areas. Different range imaging methods can be applied, such as stereo, laser sheet of light, and structured light.
This Thesis presents two applications based on range camera imaging.
The first application concerns software for a commercial body scanner. The SYMCAD system, manufactured by Telmat Industrie, France, is a range camera based on structured light. In order to get range measures, a horizontal strip pattern is projected onto the body. A video camera attached to a PC-controlled frame grabber is used for capturing an image of the body with the overlaid strip pattern. By extracting the edges of the strip pattern and defining an effective 3D representation of the body, it is possible to obtain anthropometrical measures. The software developed gives accurate results for conventional body figures and mostly also for diverging body figures such as hunchback and scoliosis. In case of failure, the software enables the operator to correct the measurement by a simple click-procedure. This work was part of the European Telematics/TIDE project EASYTEX (aesthetical, adjustable, serviceable, and mainstay textiles for disabled and elderly).
The second application concerns effective software procedures for 3D visualization. The latest evolution of 3D scanners provides highly dense 3D data clouds. Examples of application areas are architecture, archeological or industrial sites. The extreme complexity of these scenes is a problem for reading and apprehending those millions of points. In addition to 3D data, 2D color images of the scene can be registered, with both intrinsic and extrinsic calibration of the camera. This part of the Thesis presents the development of an octree data structure for an easy representation of 3D data and a visualization tool exploiting both 2D and 3D images. The work was realized in the Image & Signal Processing (TSI) group at ENST (Ecole Nationale Supérieure des Télécommunications, Paris, France) during a 18-month period as a visiting scientist.
2D image projections
octree data structure