Similarity retrieval of human motion as multi-stream time series data

It is possible to acquire human motion as 3-dimensional time series data by the development of sensing technology in recent years. The DP matching method is well known for recognition of time series data such as voice. The similarity between two time series data can be measured by DP matching. However, since human motion data consists of multiple mutually-synchronized time series data, the similarity between two human motions needs to take into consideration not only the similarities between two person´s motion data of the same part, but also two person´s degrees of synchronizations of all the part´s movements. This research aims to rate a user motion by measuring the similarity between the user motion and a teacher´s model motion. By considering two types of similarities called movement similarity and synchronization similarity, we propose a way for measuring human motion similarity