Three-dimensional hand pointing recognition using two cameras by interpolation and integration of classification scores

Abstract

In this paper, 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. In the experiment, the recognition rate was 73% at around 1 FPPI when ±10 ◦ error was allowed. Though this result was still insufficient for practical applications, we confirmed that integration of two camera information greatly improved the recognition performance.