Systematic errors in active 3D vision sensors

Active 3D vision sensors offer two main advantages over indirect means of obtaining 3D information about objects: non-reliance on ambient lighting and a higher image acquisition speed, which make them ideal for object manipulation operation in unconstrained environments. These sensors, however, can be adversely affected by the high dynamic range of the returned signal, although this can be overcome by repeated data sampling and low-pass filtering, using either a stochastic or model-driven approach. The potential for an additional drawback is investigated, namely the appearance of systematic errors in the 3D image data when the sensor´s distance to the object is within an order of magnitude of the sensor image resolution, such as might be the case in eye-in-hand robotic applications.