Generic Video-Based Motion Capture Data Retrieval

In this work we propose a novel and generic scheme for retrieval of motion capture (MoCap) data given a video query. We reconstruct skeleton animations from video clips by a convolutional neural network for 3-dimensional human pose estimation to narrow the gap between videos and MoCap data. A statistical motion signature is computed to extract both morphological and kinematic characteristics from the skeleton animations and the MoCap sequences. This as well ensures that the proposed scheme works on MoCap data with arbitrary skeleton structures. The retrieval is achieved by computing and sorting the distances between the motion signature of the query and those of the MoCap sequences which are pre-computed and stored in the MoCap database. For experimental evaluation, we respectively record a video dataset and capture a MoCap dataset with different performers, and conduct video-based MoCap data retrieval on them. Experimental results demonstrate the effectiveness of the proposed scheme.