Detecting and segmenting moving objects from a moving monocular camera is challenging in the presence of unknown camera motion, diverse object motions and complex scene structures.
Most existing methods rely on a single motion cue to perform motion segmentation, which is usually insufficient when facing different complex environments.
A novel monocular dense segmentation method is proposed, combining the strengths of deep learning and geometric model fusion methods.
The method achieves state-of-the-art motion segmentation results in a zero-shot manner, surpassing some supervised methods without training on any data.