The subject of this dissertation is object measurement in the industry by use of computer vision. In the first part of the dissertation, the project is defined in an industrial frame. The reader is introduced to Odense Steel Shipyard and its current level of automation. The presentation gives an impression of the potential of vision technology in shipbuilding. The next chapter describes different important properties of industrial vision cameras. The presentation is based on practical experience obtained during the Ph.D. project. The geometry that defines the link between the observed world and the projected image is the subject of the two next chapters. The first chapter gives a short introduction to projective algebra, which is extremely useful for modelling the image projection and the relation between more images of the same object viewed from different positions. It provides a basis for understanding many of the results later in the dissertation. In the second chapter a variety of different camera models are described. The relation between different models is explained and a guide is given to the interpretation of the model parameters. The following chapter deals with the problem of camera calibration. Different issues related to residual analysis are discussed and a calibration example is shown. The presentation is based on a software program that has been developed during the project. It is shown that the used cameras can be calibrated down to 1/20 pixel. An accurate description of the geometry is only relevant if features can be detected accurately in the images. This is the subject of the next chapter, where reference mark detection and straight edge detection are treated in two separate sections. The detection of reference marks is based on a parametric model, and it is shown that marks in synthetic images can be detected with an accuracy of 1/100 pixel. Two new methods for straight edge detection are presented. They aim at using all the image information in a global optimization. The paper ''From Hough Transform to Radon Transform using Interpolation Kernels'', which is included in a later chapter, is introduced. The new methods are compared with more conventional approaches on a number of synthetic images. It is shown that local edge detection by Gaussian convolution with a subsequent regression through the edge pixels is superior to the other methods in accuracy as well as in speed. The next chapter introduces the paper ''On Averaging Rotations'' that was presented at the 11'th Scandinavian Conference on Image Analysis, SCIA'99, in Greenland. The paper makes a theoretical comparison of two normally used linear averages to a recently proposed method that considers the structure of the non-linear manifold of rotations. The conclusion is that the behaviour of the three different methods is very similar. The following four chapters describe the practical results of the project. The first chapter gives a short introduction to a report entitled ''Reconstruction and Matching of OSS Mock-Up''. This report describes a preliminary attempt to apply a method of Euclidean reconstruction from a sequence of images on a ship block. The other three chapters describe vision installations that have been made at Odense Steel Shipyard. The first installation uses vision for check-in and quality control on a plasma cutting station. The second installation was designed for check-in on the quay. Finally, the third installation does check-in and quality control on a laser station. It is shown that measurements can be made with an accuracy of 1 mm at a distance of 10 m under favourable conditions. When the camera is only 1-2 m from the object the accuracy is better than 1/3 mm. It is also shown that it is very difficult to obtain ideal conditions and that the vision measurements are very sensitive in the large-scale installations. Only the installation on the laser cutter is currently used in production. Finally, a chapter gives some concluding remarks about the results of the project.