Analysis and Evaluation Between the First and the Second Generation of RGB-D Sensors

With the recent introduction of the new Kinect II, the second generation of the well-known Microsoft Kinect sensors, the connection between RGB-D sensors, reverse engineering, and computer vision applications is reinforced. This new sensor is based on a time-of-flight technology, which differs from the previous generation of RGB-D sensors, including other devices, such as the Asus Xtion Pro and PrimeSense Carmine, which was based on structured light. Although characterized by better technical specifications, this does not neccessarily translate to the improvements in its application tasks. This paper aims at comparing quantitatively the Kinect II with respect to the first generation of RGB-D sensors in terms of two specific application scenarios: 1) 3-D reconstruction and 2) object recognition. To this end, we propose a novel data set with ground truth obtained with a metrological laser scanner, which allows a twofold analysis: 1) a performance comparison in terms of reconstruction accuracy and 2) a comparison in terms of object recognition and 3-D pose estimation. The obtained results confirm that the new version of the Kinect sensor demonstrate higher precision and less noise under controlled conditions. Furthermore, we provide a quantitative estimation of how much such factors turn out into an improvement in terms of object recognition rate and 3-D pose estimation.