A 3D reconstruction algorithm of multiple objects

Abstract

Controlling robots based on human motion recognition visually has attracted much attention recently. Classical methods are mainly with single viewpoint, like interaction via Microsoft Kinect depth sensor. However, such methods will strictly limit the interaction range and human users have to face the visual sensor. Stereo vision methods are helpful to implement a viewpoint-free system. Visual hull, a classical stereo vision method, can provide the estimated 3D shape of the avatar and prepare further processing such as recognition. Upon the previous visual hull methods (single object), this paper proposes a visual hull algorithm which can calculate the 3D surface of multiple objects (multiple users or scene with the user and occlusion objects). When dealing with several co-existing objects, the silhouette of each object is separated first. Then the visual hull of each object is computed independently. Experimental results are shown to prove the effectiveness of our method. The proposed method will lead to a better human-robot interaction.