Crude Registration of Range Images through Matching of their Simplified Meshes

For constructing a tri-dimensional model from range images, they must be registered, i.e., it must be put together into the same coordinate system. There are good algorithms available to perform this, but only after the range images have been brought into positions that are close to the true ones with respect to an arbitrarily chosen reference system. These initial estimates are often provided manually. This paper deals with a technique to compute them automatically. We present a way to determine parameter values that are necessary to control the control flow of a simplified mesh based crude registration, which is characterized by two novel features: drastic data reduction and an expressive shape descriptor.