Real-time indoor mapping by fusion of structured light sensors

In recent years the demand for indoor mapping applications with real-time data collection has increased with the decrease in the cost of mapping sensors. In this paper, a unique, automatic system is devised to map unknown indoor environments using a combination of two optical scanners, the Kinect Scanner and the static Leap Motion Controller to collect point clouds with true color information registered to a common frame of reference. There are several innovative features of our proposed system. First, the tracking is separated from the mapping in a manner similar to outdoor mobile mapping systems i.e. there is no feedback of mapping features to aid system tracking. This facilitates the data collection mechanism immensely as the accuracy is not dependent on feature matching techniques or odometry updates required in other navigation methods. Secondly, since we use the Leap Motion Controller to track the Kinect scanner, there is no requirement for an Inertial Navigation System, and therefore no drifts accumulate in the system. This prevents erroneous loop closures that degrade the accuracy of the scanned point clouds. Thirdly, the system is immensely cheap and compact which bodes well for use in situations not demanding high-accuracy, using commercially available structured light scanners. To quantify the data collection ability of the individual sensors, calibrations are performed followed by an accuracy analysis. Fusion is performed with a description of the mathematical models that reference the separate sensors to a common coordinate system. A quantitative analysis of the fused point cloud is provided which shows that registration can be performed on a real-time basis with a point-plane accuracy of 10-11 cm that shows promise for being commercially viable option for the indoor mapping.