Study On The Method Of Human Motion Capture Data Retrieval

In recent years, multimedia technologies and3D human motion capture equipment gains rapid progress and people have accumulated a huge number of motion capture (mo-cap) data. For long, it has been proven that mo-cap data has the ability to preserve subtle details of human motion and produces realistic visual impact, so it becomes an indispensable element in animation, film making and computer games. However, facing such large scale data, how to effectively manage and reuse them becomes a new challenge. Therefore, the primary goal of this research area is to design an algorithm which is able to query and browse the mo-cap database efficiently and accurately.We first introduce the structure of the human skeleton and its mathematical representation, and then we show how to parse motion capture data from ASF/AMC files. Some existing motion retrieval methods are studied in this paper, and inspired by their advantages and disadvantages, we propose a new motion retrieval method based on Mahalanobis Distance which is learned by large margin nearest neighbor metric algorithm. The experimental results demonstrate that our method effectively improves the retrieval performance compared with the methods based on Euclidean distance and Linear Regression. The existing work we studied are listed below.1) A perceptually consistent, example-based human motion retrieval approach is implemented. The algorithm employs a motion pattern discovery and matching scheme that breaks human motions into a part-based, hierarchical motion representation. Building upon this representation, a fast string match algorithm is used for efficient runtime motion query processing.2) We implement a motion retrieval method using low-rank subspace decomposition of motion volume. The method first converts an action sequence into Self-Similarity Matrix (SSM), which is based on the notion of self-similarity. The SSMs are then used to construct order-3tensors, and we propose a low-rank decomposition scheme that allows for converting the motion sequence volumes into compact lower dimensional representations, without losing the nonlinear dynamics of the motion manifold.3) An Adaptive Feature Selection (AFS) based retrieval method is implemented, in which, AFS is used to abstract the characteristics of the query by a Linear Regression Model, and different feature subsets can be selected according to the properties of the specific query.