Matching On-vehicle Camera Image and 3D Digital Map using Features of Building Sequence

Abstract

    We reconstructed textured urban 3D models by matching in time-series a video stream of on-vehicle camera and an existing 3D digital map. For extracting features in urban scenes, we improved the conventional Epipolar Plane Image (EPI) and proposed Temporal Height Image (THI), where width and height information of building candidates are included in the space-time image. Also, we classified matching patterns of the building candidates based on the features, and designed a DP matching system. Finally, a textured 3D models could be reconstructed.