Creating accurate, current digital maps and 3-D scenes is a high priority in today’s fast changing environment. The nation’s maps are in a constant state of revision, with many alterations or new additions each day. Digital maps have become quite common.
Google maps, Mapquest and others are examples. These also have 3-D viewing capability. Many details are now included, such as the height of low bridges, in the attribute data for the objects displayed on digital maps and scenes. To expedite the updating of these datasets, they should be created autonomously, without human
intervention, from data streams. Though systems exist that attain fast, or even real-time performance mapping and reconstruction, they are typically restricted to creating sketches from the data stream, and not accurate maps or scenes. The ever increasing amount of image data available from private companies, governments and the internet, suggest the development of an automated system is of utmost importance.
The proposed framework can create 3-D views autonomously; which extends the functionality of digital mapping. The first step to creating 3-D views is to reconstruct the scene of the area to be mapped. To reconstruct a scene from heterogeneous
sources, the data has to be registered: either to each other or, preferably, to a general, absolute coordinate system. Registering an image is based on the reconstruction of the geometric relationship of the image to the coordinate system at the time of imaging. Registration is the process of determining the geometric transformation parameters of a dataset in one coordinate system, the source, with respect to the other coordinate system, the target. The advantages of fusing these datasets by registration manifests itself by the data contained in the complementary information that different modality datasets have. The complementary characteristics of these systems can be fully utilized only after successful registration of the photogrammetric and alternative data relative to a common reference frame. This research provides a novel approach
to finding registration parameters, without the explicit use of conjugate points, but using conjugate features. These features are open or closed free-form linear features, there is no need for a parametric or any other type of representation of these features
The proposed method will use different modality datasets of the same area: lidar data, image data and GIS data. There are two datasets: one from the Ohio State University and the other from San Bernardino, California.
The reconstruction of scenes from imagery and range data, using laser and radar data, has been an active research area in the fields of photogrammetry and computer vision. Automatic, or just less human intervention, would have a great impact on alleviating the “bottle-neck” that describes the current state of creating knowledge
from data. Pixels or laser points, the output of the sensor, represent a discretization of the real world. By themselves, these data points do not contain representative information. The values that are associated with them, intensity values and coordinates,
do not define an object, and thus accurate maps are not possible just from data. Data is not an end product, nor does it directly provide answers to applications, although implicitly, the information about the object in question is contained in the
data. In some form, the data from the initial data acquisition by the sensor has to be further processed to create useable information, and this information has to be combined with facts, procedures and heuristics that can be used to make inferences
for reconstruction. To reconstruct a scene perfectly, whether it is an urban or rural scene, requires prior knowledge, heuristics. Buildings are, usually, smooth surfaces and many buildings are blocky with orthogonal, straight edges and sides; streets are
smooth; vegetation is rough, with different shapes and sizes of trees, bushes. This research provides a path to fuse data from lidar, GIS and digital multispectral images and reconstructing the precise 3-D scene model, without human intervention, regardless of the type of data or features in the data. The data are initially registered to each other using GPS/INS initial positional values, then conjugate features are found in the datasets to refine the registration. The novelty of the research is that no conjugate points are necessary in the various datasets, and registration is performed without human intervention.
The proposed system uses the original lidar and GIS data and finds edges of buildings with the help of the digital images, utilizing the exterior orientation parameters to project the lidar points onto the edge extracted image/map. These edge points are then utilized to orient and locate the datasets, in a correct position with respect
to each other.