OBJECT-ORIENTED MEASUREMENT OF PIPE SYSTEMS USING EDGE MATCHING AND CSG MODELS WITH CONSTRAINTS
Johan W. H. Tangelder, George Vosselman, Frank A. Van den Heuvel · 2000
A method for the measurement of pipe systems from a collection of images is presented. A predefined library of parameterized object models defined by CSG trees is used to describe several types of piping elements. An initial measurement of each component of the pipe system is obtained by selecting the appropriate model from the library, projecting this model into the images and modifying the pose, and the shape of the model such that it is aligned with the object in the images. The images are taken with a calibrated camera. In addition, the exterior orientations of the images are assumed to be known approximately. All the selected models are combined in one tree, that describes the complete pipe system. Constraints within object models and also between object models are applied. For the precise measurement of pipe systems we developed a two stage fitting procedure. First, in a pixel based fitting stage the model edges are optimally aligned with the pixels with high gradients in a buffer around the projected model edges. Second, in an edge-model based fitting stage the grey values in the buffer are matched with the grey values predicted by a Gaussian smoothed step edge and the standard deviations of the measured parameter values are estimated. The practical application of this fitting procedure is demonstrated using object models with constraints. The use of constraints reduced the standard deviation of most of the estimated parameter values significantly. 1