Improvement of 3-D data by neural networks

The 3-D measurement by photogrammetry is easy only in special cases. The measurement principle consists in an approximation of surface pieces by planes or paraboloids. If steps in surfaces or strongly curved surfaces are measured systematic errors appear. Also random errors appear. The aim is to reduce these errors. After measuring the 3-D data will be processed by neural networks. A modified associative memory smoothes and corrects the calculated 3-D coordinates using a-priori knowledge about the measurement object. The result is the reconstructed 3-D shape. For the measurement and processing of data local coordinates are used. The calculated geometrical shape is more precise than the results obtained with other methods or needs fewer measurement values. It is also possible to measure special dimensions of parts of measured objects