On the Use of Matching Techniques for Automatic Aerial Triangulation

The matching strategy of any approach to automatic aerial triangulation (AT) plays the key role for a high success rate. It has basically to solve the problem of finding homologous image patches even in case of very crude initial approximations – and to subsequently apply a matching scheme which provides preferably fully transferred tie points. Crucial situations occur, for instance, when large relief displacements come into play or the orientation parameters are only given by a flight index map. The initialization part can be significantly simplified if the exposure centers are provided by DGPS or by a given DEM. The automation of AT has been a focus of scientific investigations in the past. Numerous groups opened the field for still ongoing developments of modules for automatic AT on today’s digital photogrammetric systems. In general, automatic AT is viewed as one of the key techniques on digital photogrammetric systems for a better overall functionality and user acceptance as well. There are great expectations that automatic AT will catapult accuracy, speed and economical efficiency to a completely new level . The solutions proposed by the scientific community or applied in operational systems mainly use the classical least-squares matching method and the feature-based matching method, however they differ in detail considerably. The paper gives a short overview about the matching strategies of several approaches. Special emphasis is given to the matching strategy of the MATCH-AT system, which has gained considerable progress recently. The integration of a block adjustment and a DEM generation technique are outstanding features which give more flexibility, especially for the matching part of the system. The characteristics of the matching strategy are discussed with results from a small sample block. Additionally, the impact of two compression techniques (JPEG, wavelet) on the accuracy of automatic AT is shown. Moderate JPEG compression rates of 1:6 deteriorate the standard deviation of object points by more than 20 %. In comparison, wavelet compression shows slightly better results.