A New Algorithm for Key Frame Xtraction Based on Depth Map Using Kinect

Abstract

In this paper, a new algorithm for key frame extraction based on depth map for hand gesture recognition is presented. The all input sequences are captured by Microsoft Kinect camera system. These methods extract three key frames from captured depth video sequence. These key frames describe dynamic gesture. The proposed extraction method is composed of two parts. The first part, labelled as space segmentation extracts the region of hand from background. The second part labelled as time segmentation splits captured sequence into three parts and marks one frame per part as the key frame. A new gesture database for evaluation of proposed method was created. The proposed method to human evaluators was compared. The experimental results show that the proposed system obtained accuracy about 90%.