Spatiotemporal Segmentation of GIS Object for Mobile Mapping System

As one fast developing field of GIS, mobile mapping system (MMS) is the trend of state of art mapping technique and digital measurable image (DMI) sequence is the main data type for MMS. Given the DMI data acquired, it's a tough but necessary task to segment the GIS object of interest (OOI) for various engineering applications. This paper proposed a novel hybrid method to segment the GIS object in DMI sequences spatiotemporally. First, the GIS OOI is indicated by the user in a key frame. We propose a hypothesis for foreground background identification and applying watershed transformation to achieve foreground background segmentation. Second, patch-based color histogram back-projection is employed to find the candidate region of the OOI in the new frame. Then SIFT is employed to describe the OOI and detect correspondence object in the new frame. The object template is updated as the matching going on until the last frame. Experimental results demonstrate the robustness and effectiveness of our method.