A Human Action Descriptor Based on Motion Coordination

Abstract

In this paper, we present a descriptor for human whole-body actions based on motion coordination. We exploit the principle, well known in neuromechanics, that humans move their joints in a coordinated fashion. Our coordination-based descriptor (CODE) is computed by two main steps. The first step is to identify the most informative joints which characterize the motion. The second step enriches the descriptor considering minimum and maximum joint velocities and the correlations between the most informative joints. In order to compute the distances between action descriptors, we propose a novel correlation-based similarity measure. The performance of CODE is tested on two public datasets, namely HDM05 and Berkeley MHAD, and compared with stateof-the-art approaches, showing recognition results.Index Terms-Gesture, human factors and human-in-the-loop, human-centered automation, posture and facial expressions.

Recurrent network based automatic detection of chronic pain protective behavior using MoCap and sEMG data

“…Additionally, for the MoCap data, we compute angles and energies, instead of using the raw Euclidean positions. This is because the angle is invariant to the change of joint position in the movement space and also to provide better representation with smaller dimensionality of body movement [32]. At each timestep, a total of 13 angles are calculated in 3D space as suggested in [8] based on the 26 anatomical points to describe the local movements of the body, where the energies are the square of their respective angular velocities.…”

Optimal Reconstruction of Human Motion From Scarce Multimodal Data

“…As mentioned above, we can identify an optimal placement of sensors by minimizing a norm of the a posteriori covariance matrix reported in (18). However, the number of DoFs to be measured must be known in advance, and this is a free parameter of the problem that has to be set as a tradeoff between estimation uncertainty and complexity of the sensing setup.…”

“…Under a pure observability point of view, i.e., for the simplification of the problem of retrieving information on human body posture and motion, the synergy-inspired dimensionality reduction has also produced interesting results [17]. In [18], the authors presented CODE, a COordination-based action DEscriptor, which considers minimum and maximum joint velocities and the correlations between the most informative joints (i.e., the joints that are mostly involved in the execution of a certain action), for whole-body action classification. In [19], a compact 6-D view-invariant skeletal feature (skeletal quad) based on a local skeleton descriptor for encoding the relative position of joint quadruples was proposed for action recognition from single depth images.…”

Learning Temporal and Bodily Attention in Protective Movement Behavior Detection

“…Following the approach in [6], each sample is characterized by 13 full-body joint-angles as well as the energies of these. Each joint angle is formed by connecting three body-joints in the 3D Cartesian space, and have the advantage, over joint positions, of being invariant to the translation and change of the reference frame [28]. The energy is the square of the respective angular velocities.…”