The segmentation of bones and bone fragments in clinical computed tomography datasets is an important first step in order to carry out computer-based surgical planning using patient-specific anatomical models. While semi-automatic and automatic methods have been proposed for the intact pelvic bone, the segmentation of bone fragments in the fractured pelvic bone still is a challenge due to weak boundaries and the diversity of injury patterns. We propose a semi-automatic multi-step segmentation method using bone and fracture gap enhancement filtering and a graph cut based bone fragment separation approach. The key contribution is a technique for automated detection of incorrect bone fragment separation in the case of incomplete pelvic fractures based on fracture gap planes detected in Hessian filtered images. Moreover, we propose a new sheetness measure based on a modified Hessian matrix. Our system is capable of segmenting fragments of complex hip fractures with only minimal user interaction.