Research On Automatic Generation Of Labanotation Based On Motion Date Segmentation

Motion capture is one technique to measure the object motion condition accurately, its data could be converted to BVH (Biovision Hierarchical Data) file by using marked or no-marked device. It could record the rotation angle of each hierarchy joint so that trajectory of action is described as in the form of data. However, it is not easy to read or show. Compared to motion data capture, Labanotation is one kind of action notation as similar as staff notation of music. Moreover, it is intuitive, convenient and easy to read. Based on questions above, we mainly study how to convert capture motion data to Labanotation automatically so that it contributes to traditional culture preservation, communication and education directly.The main work in this paper is to set up one automatic generation of BVH files to Labanotation platform. BVH file analysis, segmentation of motion capture date, gesture analysis and platform building are the main process in this paper. We set up the algorithm model and experimental platform among above process and the main contributions are as follows:(1) We set up a capture motion collection and analysis model. In this paper, we used the optical motion capture device called OptiTrack. The device can cached the motion in C3D file format, then, the motion capture data is converted into in BVH file format. Next, we used the quadruples and the euler angle to analyze the BVH files. In the specific design, we calculated the joints of hierarchy and relative rotation angle matrix to third position coordinates of 26 joints in the world coordinate system. The final motion data is a long matrix, whereby, the line number is as same as frame number and the number of rows is 78.(2) We presented an automatic segmentation approach based on beats and first frame by wavelet translate, and used the decreasing gradient algorithm to achieve subdivision and integration. With analyzing and comparing the two traditional approaches in Labanotation field, which are based on kinematic threshold and PPCA (Probabilistic PCA), we got the best one of three approaches. The feature vector of kinematic threshold segmentation is the relative position and speed between terminal node and root; the principle of PPCA is cluster algorithm after dimension reduction; the new re-clustered segmentation approach is an innovation method and it determines the wavelet scale according the motion complexity. Firstly, we used the wavelet transform to remove the noise of relative position feature, and then calculated the average distance between the peaks and valleys as the rhythm. Secondly, we got the candidate segmentation point of the motion date based on the rhythm. Thirdly, we used the decreasing gradient algorithm to calculate the local optimal based on the beat point. Then, the accurate segmentation points based on the subdivision and integration could be obtained. Compared with the three approaches, the best segmentation method for Labanotation could be selected.(3) We proposed one method of gesture analysis based on space angle division. Firstly, this model can determine the specific time’s gesture through calculating the angle between action and axis to judge the action orientation and height level. Secondly, dance symbols are generated on the third-line spectrum based on joint semantic analysis and gesture data expression.