Homography-based loss function for camera pose regression

Some recent visual-based relocalization algorithms rely on deep learning methods to perform camera pose regression from image data. This letter focuses the loss functions that embed error between two poses based regression. Existing are either difficult-to-tune multi-objective or present unstable reprojection errors ground truth 3D scene points and require a two-step training. To deal with these issues, we introduce novel function which is multiplane homography integration. new does not prior initialization only depends physically interpretable hyperparameters. Furthermore, experiments carried out well established datasets show it minimizes best mean square during training when compared existing functions.