Semantic Segmentation of Low Frame-Rate Image Sequence Using Statistical Properties of Optical Flow for Remote Exploration

For the application of well-established image analysis algorithms to low frame-rate image sequences, which are common in bio-imaging and long-distance extrapolation, we are required to up-convert the frame-rate of image sequences. For the motion analysis of low frame-rate image sequences, we introduce a two-step method for semantic segmentation of the dominant plane, which is the largest planar area on an image plane, from a low frame-rate image sequence. The algorithm first extracts candidate pixels using statistics of optical flow vectors derived by temporal optical flow super-resolution. Subsequently, the algorithm extracts a planar region by semantic labelling, accepting these candidate pixels as seed points. The minimisation of the semantic segmentation is achieved by the graph-cut method.