PERFORMANCE ANALYSIS OF TWO FITTING ALGORITHMS FOR THE MEASUREMENT OF PARAMETERISED OBJECTS

CAD models can be used to support the measurement of man-made objects. Since it is diff icult to precisely position a 3D CAD model in a 2D digital image by hand, fitting algorithms are being developed for a fast and accurate positioning. If a CAD model is approximately positioned in an image, a fitting algorithm can make use of local image gradients to fit the CAD model properly on the edges of the object in the image. In this paper two fitting algorithms are compared which differ in the estimation of the object parameters. The first algorithm examined, is the active contour model or snake algorithm minimising an energy function. The other algorithm is formulated by David Lowe and uses a least-squares adjustment. The suitabili ty of both algorithms for model-based measurements is evaluated. The speed of both algorithms is improved by using the image gradients on a grid in a buffer around the edges of the projected CAD model. A thorough performance analysis of both fitting algorithms compares the convergence behaviour, the influence of the fitting algorithm’s parameters and the precision obtained by both algorithms. From this research it appears that the least squares fitting algorithm is better suited for the semi-automatic measurement of man-made objects than the snake algorithm.