Description
It is difficult to produce pretty images for noisy point clouds and this becomes increasingly important as recent sensing techniques, for example LIDAR or Kinect, produce even more noisy input.A technique presented at Eurographics 2012, the boundary complex, yields a triangle set which approximates the shape boundary well. The number of triangles connected to vertices in the point cloud lets us determine whether those are noisy or outliers.
The point cloud is usually given as multiple range images from a sensing device. From those separate data sets the position of the sensor can be estimated and used to apply a sensor-specific noise model, since the points are mostly noisy in direction to the sensor position. We can exploit this fact to model an energy function such that both this directional noise and surface curvature are minimized.
Task
Smoothe noisy points by averaging them, remove the outlier points and fill holes using existing strategies generates a manifold boundary which is already less noisy.Determine the perspective transform from the range image as an overdetermined linear system with 6 degrees of freedom, with a standard solver. Develop a strategy to solve the above energy function efficiently.
