A New Tie Plane-Based Method for Fine Registration of Imagery and Point Cloud Dataset

Abstract

Today, both point cloud and imagery datasets processed for mapping aims. The precise fusion of both datasets is a major issue that leads to the fine registration problem. This article proposes a fine registration method based on a novel concept of tie plane. The assumption of our solution is that the laser scanner point cloud is much more accurate than the image interior and exterior geometric accuracy. In fact, we register the inaccurate image network to the accurate point cloud data. To do this, tie points are extracted from images. Then, the fine registration is commenced by filtering the unstable tie points as the preprocessing phase. Subsequently, tie planes are reconstructed around the remaining tie points by photogrammetric space intersection. The tie planes are locally fitted to the point cloud data via both normal and directional vectors. Afterward, a novel combined bundle adjustment is developed based on the conventional tie point equations and the new tie plane constraints. Therefore, the interior and exterior orientation parameters are refined. To evaluate our solution, both indoor and outdoor datasets are experimented. The results illustrate a registration error of about <1.6 pixels for both datasets, indicating ∼23% to 40% average accuracy improvement compared to the existing methods.