Three-Dimensional Hand Pointing Recognition using Two Cameras

We propose a novel method of hand recognition for remote mid-air pointing operation. In the proposed method, classification scores are calculated in a sliding window for hand postures with different pointing directions. Detection of a pointing hand and estimation of the pointing direction is performed by interpolating the classification scores. Moreover, we introduce two cameras and improve the recognition accuracy by integrating the classification scores obtained from two camera images. The result of the experiment showed that correct hand detection and direction estimation were realized, and that the accuracy was greatly improved by integrating two camera information.
Dai Fujita, Takashi Komuro: Three-Dimensional Hand Pointing Recognition using Two Cameras by Interpolation and Integration of Classification Scores, Proc. 13th European Conference on Computer Vision (ECCV 2014) workshops (2014) / Lecture Notes in Computer Science (LNCS), Vol. 8925 (Computer Vision - ECCV 2014 Workshops), pp 713-726 (2015)