Recent work has shown that flow estimation from a pair of images can be formulated as a supervised learning task to be resolved with convolutional neural networks (CNN). However, the basic straightforward CNN methods estimate optical flow with motion and occlusion boundary blur. To tackle this problem, we propose a tiny diagnostic dataset called FlowClevr to quickly evaluate various modules that can use to enhance standard CNN architectures. Based on the experiments of the FlowClevr dataset, we find that a deformable module can improve model prediction accuracy by around 30% to 100% in most tasks and more significantly reduce boundary blur. Based on these results, we are able to design modifications to various existing network architectures improving their performance. Compared with the original model, the model with the deformable module clearly reduces boundary blur and achieves a large improvement on the MPI sintel dataset, an omni-directional stereo (ODS) and a novel omni-directional optical flow dataset.

From the viewpoint of applications, conventional cameras are usually limited in their fields of view. The omni-directional camera has a full range in all directions, which gains the complete field of view. In the past, a moving object can be detected, only when the camera is static or moving with a known speed. If those methods are employed to mobile robots or vehicles, it will be difficult to determine the motion of moving objects observed by the camera. In this paper, we assume the omni-directional camera is mounted on a moving platform, which travels with a planar motion. The region of floor in the omni-directional image and the brightness constraint equation are applied to estimate the ego-motion. The depth information is acquired from the floor image to solve the problem that cannot be obtained by single camera systems. Using the estimated ego-motion, the optical flow caused by the floor motion can be computed. Therefore, comparing its direction with the direction of the optical flow on the image leads to detection of a moving object. Due to the depth information, even if the camera is in the condition that combining translational and rotational motions, a moving object can still be accurately identified.