Abstract: The demand of 3D model retrieval is increasing, and the sketch-based method has been
proved to be the most effective and efficient approach to retrieve 3D models. In order to get better
experience on sketch retrieval, it’s necessary to focus on the way of human-computer interaction as
well as improve it. The Leap Motion sensor can detect the position information of fingers and
gestures with high precision in real-time, which makes volume interaction with large screen at
long-distance more effective. In this paper, an extended manifold ranking method based on Leap
Motion is presented as a new retrieval framework. The sketch drawing and model retrieval processes
are realized by obtaining data of hands from Leap Motion to the browser. In the process of sketch
retrieval, manifold ranking is proposed to rank on similarities between models combining with
semantic classification as a constraint, thus improving the ranking score of models belong to the same
classification. Experiments have demonstrated that the proposed method makes the sketch retrieval
more closable and convenient to the user as well as improves the accuracy of 3D model retrieval.