Deformable Shape-From-Motion in Laparoscopy using a Rigid Sliding Window

We present an approach for deformable 3D surface reconstruction using laparoscopic monocular video data that is based on breaking down the nonrigid reconstruction problem into a sequence of temporally-localised rigid reconstructions. Our main contribution, the Rigid Sliding Window algorithm provides a means to achieve this whereby Rigid Shapefrom-Motion is performed over a sliding window that adapts according to the amount of deformation detected in the scene. Our method is particularly applicable for laparoscopy, where, by adapting the window, it can exploit low-frequency deformations caused by for example breathing. Our algorithm is demonstrated on both synthetic and in-vivo image sequences.