The Microsoft Kinect sensor is getting prominent in several
applications beyond the original field of videogames. The main
advantage of this 3D measurement device is the low price and
easy software development. Kinect has a Flash LiDAR type sensor
and RGB camera to provide colored 3D point cloud acquisition.
Based on previous accuracy investigations, the range calibration
results were found rather good for a low-end device. The
accuracy, however, can be further increased with more
sophisticated sensor calibration, especially by using the IR
intensity values for co-registration, as the number of matching
points between the depth and RGB images can be further
increased, which may yield to better calibration performance.
Previous experiences indicated some object distance dependent
scaling issues for depth calculation; objects farther than 2
meters are scaled down, caused by the measurement quantization
required for data transmission. In addition, the spacing of the
points is growing with the object distance. This means that the
common sphere fitting methods fail with Kinect on farther
objects. Therefore, a new fitting method was developed for
sphere fitting on Kinect acquired point cloud, which yields
better results, even over 2 meters. This way, the raw
measurement values can have high accuracy and, in particular, if
some preliminary information about the object morphology exists.
