Movement data gesture classification process based on DTW curve

Abstract

The invention discloses a method for classifying motion data actions based on DTW curved line. The method comprises the following steps: (1), a bidirectional DTW distance and a partitioned DTW distance which are used for measuring similarities among movement sequences are produced; (2), a DTW curved line is produced according to the bidirectional DTW distance and the partitioned DTW distance, wherein, W EoLC is a logical division validity curved line, D s-dtw is a partitioned DTW distance, lambada is a threshold value used for partitioning, and P and Q represent two movement sequences; and (3), hierarchical clustering algorithm is adopted for classifying the movement sequences according to the valid weighted distance obtained from step (2). The method of the invention relaxes the restriction on DTW, so that the classification robustness is enhanced, and experimental results show that the method of the invention can ensure that better classification results can be obtained.

Description

technical field [0001] The invention relates to a method for automatic classification of motion data, which is mainly used for action classification. Background technique [0002] In order to effectively use 3D motion data (such as training data for action recognition), it is necessary to divide action segments with the same motion characteristics into one category, and then label each category to form various action sets. The set of action sets is called for the motion library. For example, a large number of "walking" and "running" action segments are classified and labeled to form two action sets of "walking" and "running". The existing classification methods mainly include: [0003] (1) Manual classification method, that is, using manual classification and labeling of motion data, this method will consume a lot of time and energy; [0004] (2) The manifold clustering method Manifold Clustering proposed by Souvenir (Souvenir R., Pless R. Manifold Clustering [A]. Proceed...

AI Technical Summary

Problems solved by technology

[0003] (1) Manual classification method, that is, using manual classification and labeling of motion data, this method will consume a lot of time and energy;

[0004] (2) The manifold clustering method Manifold Clustering proposed by Souvenir (Souvenir R., Pless R. Manifold Clustering [A]. Proceedings of the Tenth IEEE International Conference on Computer Vision (ICCV'05) [C]. Washington: IEEE Computer Society , 2005: 648-653), this method can only classify the motion data of simple behavior according to the number of clusters manually specified;

But for action classification, these conditions are too strict and perform poorly in logical judgment

[0010] (2) Poor robustness: the DTW distance of two sequences is a specific value, and only one value is used to judge the similarity of two sequences, which has poor robustness and is susceptible to noise interference

Images