Hand position tracking using a depth image from a RGB-d camera

Three algorithms for hand position tracking are presented. These algorithms work in real time, have low computational cost and only use the depth image obtained from a RGB-d camera, therefore they are light and skin color invariant. Despite the fact that there are libraries that perform hand position tracking using RGB-d cameras (Like Microsoft Kinect SDK, and PrimeSense's NITE), these libraries generally do not have their algorithms documented. The algorithms presented in this paper were developed with the purpose of providing a set of well documented algorithms so improves can be proposed. The algorithm with the best performance runs between 7.1ms and 3.4ms, with an error of 17 mm. The algorithms can be used for natural user interfaces, they have been used for the guidance of the end effector of an industrial robot; they were also used for hand segmentation which is commonly the input for full hand pose estimation.