Localization System using Microsoft Kinect for Indoor Structures

Abstract

This paper proposes a localization system using the Microsoft Kinect sensor. It is difficult to accurately measure self-position and self-posture by a red-green-blue (RGB) image sensor, since the accuracy in the depth direction of an image sensor tends to be worse than that in other directions using only an RGB image sensor. The Kinect sensor has both an RGB image sensor and a depth sensor. This sensor can directly calculate the depth value; therefore, the accuracy in the depth direction is better. In the proposed system, natural feature points are detected from an image by a Harris interest operator, and the depth data of these feature points are obtained from the depth sensor. After that, these points are tracked by the template matching method. The camera's position and posture are calculated from the position of these tracked points. Finally, we provide examples of 3D scene reconstruction and estimation results.