Motion reconstruction using sparse accelerometer data

Abstract

The development of methods and tools for the generation of visually appealing motion sequences using prerecorded motion capture data has become an important research area in computer animation. In particular, data-driven approaches have been used for reconstructing high-dimensional motion sequences from low-dimensional control signals. In this article, we contribute to this strand of research by introducing a novel framework for generating full-body animations controlled by only four 3D accelerometers that are attached to the extremities of a human actor. Our approach relies on a knowledge base that consists of a large number of motion clips obtained from marker-based motion capturing. Based on the sparse accelerometer input a cross-domain retrieval procedure is applied to build up a lazy neighborhood graph in an online fashion. This graph structure points to suitable motion fragments in the knowledge base, which are then used in the reconstruction step. Supported by a kd-tree index structure, our procedure scales to even large datasets consisting of millions of frames. Our combined approach allows for reconstructing visually plausible continuous motion streams, even J. Tautges and T. Helten were financially supported by grants from Deutsche Forschungsgemeinschaft (WE 1945/5-1 and MU 2686/3-1).

Filtered pose graph for efficient kinect pose reconstruction

“…Such a motion prior concept is applied to real-time pose reconstruction by generating higher quality movements [25]. When adapting this idea to accelerometer-based systems, an online lazy neighbourhood graph is used to minimize false positive samples in the local subspace [38]. When applying these methods to reconstruct Kinect poses, the main problem is that they assume the input data to be accurate, whereas the joint positions estimated by Kinect are noisy or even incorrect due to sensor error and occlusion.…”

Retrieval, Recognition and Reconstruction of Quadruped Motions

“…These sensors have been used to reconstruct full body poses of human motions (Tautges et al, 2011) from a sparse sensor setup. The feature sets developed on the basis of three dimensional information are described in Table 1.…”

Motion Tracking, Retrieval and 3D Reconstruction from Video

“…They employ 15-dimensional feature sets based on positions of hands, feet and head for fast similarity search. Later on, Tautges et al [1] enhance the LNG into incremental online version named online lazy neighbourhood graph (OLNG) and reconstruct human motions using sparse accelerometer data. They reconstruct human motions with the help of data-driven prior model which measures a-priori likelihood of the input motion into MoCap database.…”

“…After getting suitable feature sets both from input signals as well as from motion capture database, we are able to perform efficient similarity search and retrieve nearest neighbours from database. For that purpose, we have developed a kd-tree data structure and the so called online lazy neighbourhood graph (OLNG) along the lines of Tautges et al [1]. In our domain, we have adapted these methods to work with 2D feature sets extracted either from MoCap data or video data used as control input signal.…”