In this paper, we present a new approach for coarse segmentation of tubular anatomical structures in 3D image data. Our approach can be used to initialise complex deformable models and is based on an extension of the randomized Hough transform (RHT), a robust method for low-dimensional parametric object detection. In combination with a discrete Kalman filter, the object is tracked through 3D space. Our extensions to the RHT feature adaptive selection of the sample size, expectation-dependent weighting of the input data, and a novel 3D parameterisation for straight elliptical cylinders. For initialisation, only little user interaction is necessary. Experimental results obtained for 3D synthetic as well as for 3D medical images demonstrate the robustness of our approach w.r.t. image noise. We present the successful segmentation of tubular anatomical structures such as the aortic arc or the spinal chord.