3D reconstruction with shape of silhouettes
The task is to reconstruct 3D object from multiple calibrated images by implementing a naive silhouette extraction algorithm. Images and camera calibrations are provided.
Extract silhouettes from the provided images, using a simple thresholding technique. Adjust the silhoutteThreshold so that the silhouettes of the statue are clearly extracted.
Define the volume of interest. At first use a larger bounding box than necessary, just to make sure the statue has been included. Then make the bounding box tight to get better resolution. Similarly, we can start with a coarse grid, say 10 * 10 * 20, and then try a larger grid, at least 64 * 64 * 128.
Compute the occupancy score at each voxel. For each voxel, transform the point from volume to world coordinates using the transformation provided. Then project the points into the image. If a projected point falls within the silhouette, add 1 to the score.