Octree segmentation based calling gesture recognition for elderly care robot

Abstract

This paper presents a method of calling gesture recognition by isolating the head and hand of a caller based on octree segmentation. The recognition of calling gestures is designed here mainly for elderly to call a service robot for their service request. A big challenge to solve is how to make the calling gesture recognition work in a complex environment with crowded people, cluttered and randomly moving objects, as well as illumination variations. The approach taken here is to segment out individual people from the 3D point cloud acquired by Microsoft Kinect or ASUS Xtion Pro and detect their heads and hands in certain geometric configurations. The segmentation is done fast by representing the 3D point cloud in octree cells and clustering those octree cells connected by the neighborhood relationship. The head and hand in a certain geometric configuration are identified from the candidate regions defined with a segmented object and by detecting the shape and color evidences. Color model in HSV color space also discussed to well define the skin color model. The proposed method has been implemented and tested on HomeMate, a service robot developed for elderly care. The result of performance evaluation is given.