Karel, Dorninger, et al. 109 IN SITU DETERMINATION OF RANGE CAMERA QUALITY PARAMETERS BY SEGMENTATION Wilfried KAREL 1 , Peter DORNINGER 2 , Norbert PFEIFER 2 1 Christian Doppler Laboratory for “Spatial Data from Laser Scanning and Remote Sensing” at the Institute of Photogrammetry and Remote Sensing (IPF) Vienna University of Technology, Austria wk@ipf.tuwien.ac.at 2 Institute of Photogrammetry and Remote Sensing (IPF) Vienna University of Technology, Austria {pdo,np}@ipf.tuwien.ac.at KEY WORDS: Range Imaging, Range Camera, Quality, Segmentation, In Situ. ABSTRACT The present article documents the derivation of quality parameters from raw range imaging data captured under simplified real-world conditions. A sequence of thousand images showing cuboids from constantly changing points of view and in different camera attitudes serves for evaluation. The data are assigned to plane segments by means of a highly robust segmentation algorithm, and without user interaction. The according residuals are evaluated statistically with respect to known error factors. These are the observed distance, the signal intensity, the position in the field of view, and the angle of incidence. 1 INTRODUCTION Range Imaging (RIM) augments the variety of optical 3-d measurement techniques being a rather new method with remarkable characteristics. A bundle of distances is determined simultaneously at every pixel of a two-dimensional sensor array of a range camera. These ranges are deduced from the time it takes an emitted signal to return to the device. As opposed to the self-evident approach to register the time-of-flight directly using light pulses, the indirect way employing continuously modulated signals has already achieved a feasible development state. By sampling the return at every quarter of the modulation period, the phase shift and hence the object distance are deduced. RIM combines and supplements properties of other commonly used optical 3-d measurement techniques (i.e. digital imaging and laser scanning). A unique benefit can be stated concerning the registration also referred to as the reconstruction of the exterior orientation. Contrary to interferometry and triangulation by image matching or projected light patterns, RIM does not require multiple observation rays to determine a point’s position and thus minimizes the number of device setups necessary to capture complex objects. Unlike laser scanning, the area-based gathering of point data by range cameras allows for their application to kinematical scenes and for a mobile utilization without the demand for inertial and / or positioning devices. The compact dimensions of range cameras further increase their fitness for mobile use. Moreover, RIM supersedes the need for time-consuming image correlations. Holding an array sensor featuring frame rates of some fifty Hertz, RIM offers high data rates. Finally, the purchase costs of range cameras are comparable to the ones of digital amateur SLR cameras and thus far below those of laser scanners. Besides these promising advantages, severe drawbacks have to be mentioned for current RIM systems. In addition to the partially proved low temporal instrument stability, primarily the distance measurement precision and accuracy are still limited due to large systematic and random errors.