In the context of the female pelvic medicine, non-invasive magnetic resonance imaging is widely used for the diagnosis of pelvic floor disorders. Nowadays, in the clinical routine, diagnoses rely largely on human interpretation of medical images, on the experience of physicians, with sometimes subjective interpretations. Hence, image correlation methods would be an alternative way to assist physicians to provide more objective analyses with standard procedures and parametrisation for patient-specific cases. Moreover, the main symptoms of pelvic system pathologies are abnormal mobilities. The finite element model simulation is a powerful tool for understanding such mobilities. Both the patient-specific simulation and the image analysis require accurate and smooth geometries of the pelvic organs. This paper introduces a new method that can be classified as a model-to-image correlation approach. The method performs fast semi-automatic detection of the bladder, vagina and rectum from magnetic resonance images for geometries reconstruction and further study of the mobilities. The approach consists of fitting a B-spline model to the organ shapes in real images via a generated virtual image. We provided efficient, adaptive and consistent segmentation on a dataset of 19 patient images (healthy and pathological).