Matching topographic surfaces: Application to lidar and photogrammetric surfaces

If photogrammetry has been used for a long time to describe the topography, airborne laser systems are nowadays well-known to provide an accurate representation of terrestrial landscapes through irregular 3D point clouds. Both technologies have their pros and cons that a joint use may optimize to reach a better description of 3D scenes. Beyond the adjustment problem of laser strips, combining optical and laser data is at first a registration problem, especially when using high resolution images. We propose in this paper a methodology for registering laser strips with regard to a photogrammetric derived Digital Surface Model (DSM) which has been computed from a set of known pose calibrated images. Based on the main hypothesis that the geometrical frames of laser systems and digital cameras are linked with a regular function, we describe an algorithm based on the calculation of linear approximations of this transform with 3D local translations. Due to the irregular spatial distribution of laser data and the difficulty of detecting homologous points in the DSM, we adopt a statistical strategy to find patch correspondences which leads to analyze the distribution of a certain vector set of potential homologous candidates. We then propose to estimate an analytical model through a weighted sliding window strategy. Laser strips are corrected globally in a chronological order. The algorithm is validated onto synthetic transforms. Experimental results show the capability of the registration algorithm on raw laser data sets.