| Interactive Modeling: Improvement of the Reconstruction Process of a 3D Model through Real-Time Compression |
| Masters |
| Jean-Daniel Deschênes |
| Patrick Hébert (Supervisor) |
|
| Problem: The process of classical 3D modeling consists in generating a numerical model of the geometry of an object’s surface. The model is encoded with the help of a representation generally in the form of a cloud of points, a triangulation or an implicit function. This process involves two distinct steps: data acquisition and the generation of the model. If the model obtained is incomplete or unsatisfactory, the required data must be added and a new model must be generated. This latter step is usually long and delays the process of obtaining the desired model. The current objective in 3D modeling is to generate the model of the object simultaneously with the data acquisition process. Thus the problem evoked above will be solved in that the user can evaluate the quality of the model during the acquisition. This type of modeling is referred to as interactive modeling. |
| Motivation: In interactive modeling, the complexity of the algorithms used to generate the model must be linear with the quantity of data acquired. If this constraint is not respected, no computer will be powerful enough to maintain the interactivity of the system during a continuous and prolonged acquisition. Over the past few years, the 3D Vision group of the Computer Vision and Systems Laboratory has devoted considerable efforts towards the development of a modeling chain respecting the constraint of linearity. This modeling chain involves a new representation based on vector fields. The development of algorithms adapted to this representation has reached sufficient maturity to enable the modeling of a variety of objects. However, the experience acquired brought to light a limitation of the representation: the great need for memory space. This characteristic limits the size and resolution of the objects which can currently be modeled. |
| Approach: My research work aims to reach beyond this limit by developing a real time compression algorithm which can be integrated into the existing modeling chain. The algorithm will automatically choose the “optimal” level of resolution to encode the surface during acquisition. This optimal resolution level will be defined locally as a function of the complexity (level of detail) of the object and the precision of the sensor. Thus a large region with few details may require only a small quantity of memory, which represents a significant improvement over currently used methods. The newly developed algorithm must also respect the constraint of linear complexity. |
| |
| |
| |
| Calendar: September 2005 - September 2007 |
| |
| |
| Last modification: 2007/10/01 by desche07 |
