Evaluating SLAM Approaches for Microsoft Kinect

Corina Schindhelm
unpublished
The weak performances of GPS within buildings is the reason for a lot of different approaches for indoor positioning, e.g., by using WiFi or odometry. The current position of a person is crucial, for example, for location based services that increase not only outside of buildings. Navigation systems in subway stations is just one obvious example, where GPS fails to deliver the necessary information. Especially in the field of visual odometry, there are many approaches. But all of them are
more » ... l of them are either based on normal 2d camera systems or on expensive 3d camera systems. In the presented approach, we use a Microsoft Kinect, as these systems are inexpensive and widespread. We evaluate how different state of the art techniques like RANSAC or ICP can be used in combination with the Kinect and how they perform in different indoor scenarios. Our evaluation shows that those techniques can be used for the Kinect but have their shortcomings in different scenarios. For that reason, a hybrid technique was developed which combines those methods using a Kinect specialized ICP weight function. In addition, we use a loop detection algorithm to further optimize the accuracy. Finally, we present our results obtained during tests in three different test environments. This paper presents the result of different SLAM approaches implemented on the Microsoft Kinect in order to calculate trajectories.
fatcat:ystidegtvvha3fo2v4rrpikqfm