One approach in the field of crime scene documentation used the optical digitizer ��Konica-Minolta Vivid 910�� in different sensor configurations to generate a multi-resolution map [8]. This triangulation-based range sensor works within a range between 0.6 m and 2.5 m with an accuracy up to a tenth of a millimeter [13] and can be equipped with different optics (wide, middle, tele) to acquire point-clouds in different resolution classes. Using these optics, a simulated crime scene was imaged from multiple viewpoints, and various point clouds with different resolutions were recorded. These point-clouds were aligned using the semiautomatic procedure ��ImAlign�� from the commercial PolyWorks software.Nevertheless, most of the approaches required the interaction of an operator for the alignment of single 3D point-clouds or the fusion of 3D range and 2D image data.

Locating and aligning 3D-models to a scene containing multiple different objects is a well-known problem in computer vision. So-called 3D keypoint detectors [14,15] are used to generate and describe a set of distinct points of the model and all points of the scene. Thus, homologous keypoints of the model and the scene can be used to calculate the 3D-transformation matrix between both datasets.A further application where automated object recognition is required is the well-known ��bin-picking problem�� in robotics. The goal of the bin-picking approach is the automated interaction of a robot with its direct environment. Therefore, objects that should be picked up by a robot have to be identified in a 2D or 3D image.

The recognition step is commonly realized by using different feature descriptors, like, e.g., scale-invariant feature transform (SIFT) [16] or fast directional chamfer matching (FDCM) [17] in the 2D case and, e.g., the RANSAM (random sample matching) algorithm [18] in the 3D case.Using established commercial sensors for data acquisition is quite expensive and requires Batimastat intensive operator training. With the rise of different low-cost 3D sensors, an economic and simple imaging alternative is available. These sensors usually work within the same accuracy range as the more expensive ones, but with the advantage of much lower investment costs (Table 1) [19]. Furthermore, the technology of low-cost systems, which are based on consumer products, is commonly very user-friendly; thus, they are usable without intensive training.Table 1.Overview of the main sensor properties for the David and the Kinect sensor.In this study, we present an automated alignment approach for 3D point-clouds. This approach combines the advantages of multiple sensors regarding measuring volume and resolution by generating a multi-resolution map.