Uncalibrated Camera Vision Pointing Recognition for HCI

Abstract

Among gestures in non-verbal communication, pointing gesture can be taken as one of natural human computer interfaces. Vision based hand pointing is an optimal model for human-computer interaction (HCI). One of key problems among the vision based pointing gesture is how to recognize the pointing. Aiming at some limits existing in the literature, a novel method is developed to estimate pointing gestures based on some non-calibrated cameras. Multiple un-calibrated cameras are adopted to determine the pointing target based on pointing features extracted from multiple cameras and support vector machine (SVM) classifier. No explicit constraints are set on the cameras placement. Pointing user can move freely inside a wider interaction environment while pointing at some targets. The mentioned approach does not constrain the pointing surface whether is flat or not, or the target is visible by the cameras. Edge detection based on multi-scale wavelet transformation is used to extract pointing objects from a clutter background. Experiments have shown that the developed approach is efficient for pointing recognition by comparisons.