Summary
In this chapter, we discussed how Active Appearance Models can be combined with the POSIT algorithm in order to obtain a 3D head pose. An overview on how to create, train, and manipulate AAMs has been given, and you can use this background for any other field, such as medical, imaging, or industry. Besides dealing with AAMs, we got familiar with Delaunay subdivisions and learned how to use such an interesting structure as a triangulated mesh. We also showed you how to perform texture mapping in the triangles using OpenCV functions. Another interesting topic was approached in AAM fitting. Although only the inverse compositional project-out algorithm was described, we could easily obtain the results of years of research by simply using its output.
After enough theory and practice of AAMs, we dived into the details of POSIT in order to couple 2D measurements to 3D ones, explaining how to fit a 3D model using matchings between model points. We concluded the chapter by showing how to use...
