Incremental construction of 3D models of objects

The project goal is to develop and implement a real-time incremental 3D body reconstruction system that combines visual features and shape-based alignment using a Kinect(tm) depth sensor and video cameras s. Methodologies introduced previously are expected to be implemented targeting real-time performances by exploring the implementation of incremental reconstruction algorithms in a GPU to obtain 3D mesh models in realtime. Description of work 3D modelling consists mainly on four main phases: scan object surface from different views, register the views and integrate these views. The work will focus on last two phases aiming to render the reconstructed body model in mixed reality space.

Work plan:

  1. Analyze the state of the art. Perform a literature survey about efficient methodologies for pose tracking, view integration and rendering.
  2. Adapt a rotating platform to be controllable from the acquisition software using an Arduino board.
  3. Develop the software to perform the scanning of 3D objects by controlling their orientation of the rotating platform.
  4. Compute 3D mesh from dense depth maps in order lower the data to be processed and create a 3D mesh representation. Mesh simplification. Implementation of efficient algorithms (C++, OpenGL ES) for mesh surface extraction and mesh integration: an incremental version of the Crust Algorithm, and incremental integration model algorithm based on confidence measures.
  5. Adaptation for self-moving targets, such as persons.

Supervision: Prof. Paulo Menezes & MSc. Luís Almeida