Querying 3D Mesh Sequences for Human Action Retrieval

Abstract

In this paper, the unsupervised human action retrieval problem in 3D mesh sequences is addressed. An action is represented by a mesh sequence wherein each frame is represented by a static shape descriptor. Six state-of-the-art static descriptors are used to extract meaningful information of the mesh objects in the sequence. These descriptors are first examined in terms of similarity performance using Receiver Operating Characteristic (ROC) curves. Then, they are utilized in the action retrieval problem, where the query is a 3D mesh sequence. Each descriptor for an action, is considered as a multidimensional curve which traverses the corresponding points. The temporal similarity estimation is achieved by two variations of the descriptor's normalization in the context of Multidimensional Dynamic Time Warping. Experimental results are given for two standard datasets, one containing real data and another one containing artificial data.