SEMANTIC FEATURE BASED REGISTRATION OF TERRESTRIAL POINT CLOUDS

As an active remote sensing technique, Terrestrial Laser Scanning (TLS) is popular for constructing detailed 3D models of complex objects. To create a complete 3D scene, TLS point clouds scanned from multiple positions need to be registered to the same coordinate system. Conventional registration methods demand significant amount of manual work. Recent studies use geometric features (points, lines and planes) matching to automate the registration procedure. However, a fully automatic and simple registration method is still a popular research interest. In this paper, a new registration approach is proposed which uses semantic information to automate the geometric feature matching. Knowledge is used to identify semantic meaning of geometric features and classify them. The semantic type and spatial pattern of features is organized and analysed in a hierarchical manner. This provides a basis for structured feature matching reducing the search space and hence instigating automation in registration. Geometrical properties of obtained matches are used for coarse and fine registration through least squares adjustment. Our proposed approach was tested for registering building façade scans which yielded successful registration with high accuracy level