Open ddetone opened 11 years ago
First, the map is initialized via the 5-point stereo algorithm + camera translation. This generates a set of typically 200-400 3D map points. Then, within MapMaker, RANSAC is run to determine a set of inliners from every point in the map which minimize the distance from the map point to the predicted plane. From these inliners, the plane is defined by the mean position and principal component with minimal variance.
Knowing this, we should be able to use mouse input to specify a specific set of map points (rather than the entire set, which is done in initialization) to use in the RANSAC plane estimation. We can then use this estimated plane as the basis for multiple AR canvases.
Breakout:
Figure out what techniques are being used to choose the initial plane that is drawn