Performance Analysis of Kinect Sensor Trajectory Reconstruction

C.K. Toth, G. Jozkow, Z. Koppanyi, D. Grejner-Brzezinska

Abstract:	The Microsoft Kinect™ sensor has gained popularity in a large number of applications beyond its intended original design of being a 3D human interface device, including indoor mapping and navigation of pushcart and backpack sensor platforms. Indoor mapping and personal navigation systems are generally based on a multisensory integration model, as currently no sensor itself can provide a robust and accurate solution. To assess the error budget as well as to support the design of such systems, the individual sensor error budgets should be known (estimated). In this paper, a performance analysis of the Kinect sensor is provided based on a series of indoor tests, where sufficient control was available. The main goal of the study is to assess the trajectory reconstruction performance from Kinect imagery only; note that only widely available mainstream imaging tools are used. The investigation aims at estimating and evaluating the total error budget of the 3D mapping process that is based on simultaneously using RGB (2D) and depth (3D) images. The overall error budget is divided into two main parts: (1) the sensor error budget, and (2) the object space error contribution. The first part defines a lower error bound for the 3D object space observation estimation errors, i.e. what can be achieved under ideal conditions. The second part is about the object space dependency, that is the error introduced by the scene content in terms of geometry and texture that can be exploited to identify matching features in the image sequence. While it is difficult to encapsulate the impact of the object space in a rigorous sense, tendencies can be identified based on statistical evaluation of data acquired under typical object space scenarios.
Published in:	Proceedings of the ION 2013 Pacific PNT Meeting April 23 - 25, 2013 Marriott Waikiki Beach Resort & Spa Honolulu, Hawaii
Pages:	542 - 550
Cite this article:	Toth, C.K., Jozkow, G., Koppanyi, Z., Grejner-Brzezinska, D., "Performance Analysis of Kinect Sensor Trajectory Reconstruction," Proceedings of the ION 2013 Pacific PNT Meeting, Honolulu, Hawaii, April 2013, pp. 542-550.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In