Human Motion Retrieval Based On Hand-drawn Sketch

With the development of motion capture techniques,motion capture data has been widely used in various fields,such as special effects, video games and virtual reality. However,the process of human motion capture is very time consuming and expensive. So,it is important to reuse the pre-captured human motion in database. To this end, an easy,efficient and accurate method to retrieve motion from the large amount of recorded motion capture data is extremely necessary.In this paper, a sketch-based human motion retrieval method is proposed. With some simple rules, the user can acquire a desired motion by sketching several key postures. To deal with the connection between 2D skeching key posture and 3D motion in the database,the key postures of the 3D motions are projected onto 2D planes, then an effective feature is extracted from these projeced-postures. Furthermore, a novel index structure based on k-d tree is constructed to index the features, which greatly speeds up the retrieval process. With our posture-by-posture retrieval algorithm and the motion synthesis technique based on motion graph, the desired motion can be got accurately.Two feature expression methods are proposed in this paper:Limb Direction(LD) feature and 2D Geometric Posture Descriptor(2GPD). The two features are very effective and efficient to rep-resent both 2D sketching posture and projected-posture. For the high dimension 2GPD feature, a semi-supervised feature selection algorithm derived from Laplacian Score is adopted to select the most discriminative feature subset.In conclusion, the sketch-based input mode simplifies user’s operation. The effectiveness of the feature and the use of the index structure promote efficiency of retrieval. The posture-by-posture retrieval algorithm and the motion synthesis technique based on motion graph pledge retrieval ac-curately.