Search

Establishing a human motion reconstruction system from very few constraints imposed on the body has been an interesting and important research topic because it significantly reduces the degrees of freedom to be managed. However, it is a well-known mathematically ill-posed problem as the dimension of constraints is much lower than that of the human pose to be determined. Therefore, it is challenging to directly reconstruct the whole body joint information from very few constraints due to many possible solutions. To address this issue, we present a novel deep learning framework with an attention mechanism using large-scale motion capture (mocap) data for mapping very few user-defined constraints into the human motion as realistically as possible. Our system is built upon the attention networks for looking back further to achieve better results. Experimental results show that our network model is capable of producing more accurate results compared with previous approaches. We also conducted several experiments to test all possible combinations of the features extracted from the mocap data, and found the best feature combination to generate high-quality poses.

The inertial measurement unit (IMU) and magnetic, angular rate, and gravity (MARG) sensor orientation and position are widely used in the medical, robotics, and other fields. In general, the orientations can be defined by the integration of angular velocity data, and the positions are also computed from the double integration of acceleration data. However, the acceleration and angular velocity data are often inaccurate due to measurement errors which arise when the sensor moves quickly. Therefore, the orientations and positions significantly differ from the actual values. To address these issues, several techniques are proposed for the accurate measurement of IMU and MARG sensor orientations and positions. The proposed optimization method is applied to raw sensor data to compute faithful orientations by stabilizing and accelerating the convergence of the optimization process. Furthermore, a deep neural network based on 1D convolutional neural network (CNN) layers is proposed to predict the desired velocity from raw acceleration data. The method is validated qualitatively and quantitatively with an optical motion capture (mocap) system. The experimental results show that the proposed method significantly improves orientation and position estimations compared to those of other approaches.

In this paper, we propose a robust low-cost mocap system (mocap) with sparse sensors. Although the sensor with an accelerometer, magnetometer, and gyroscope is cost-effective and offers the measured positions and rotations from these devices, it potentially suffers from noise, drift, and lost issues over time. The resulting character obtained from a sensor-based low-cost mocap system is thus generally not satisfactory. We address these issues by using a novel deep learning framework that consists of two networks, a motion estimator and a sensor data generator. When the aforementioned issues occur, the motion estimator feeds the newly synthesized sensor data obtained with the measured and predicted data from the sensor data generator until the issues have been resolved. Otherwise, the motion estimator receives the measured sensor data to accurately and continuously reconstruct the new character poses. In our examples, we show that our system outperforms the previous approach without the sensor data generator and we believe that it can be considered a handy and robust mocap system.

In this paper, we propose a novel method of denoising human motion using a bidirectional recurrent neural network (BRNN) with an attention mechanism. The corrupted motion that is captured from a single 3D depth sensor camera is automatically fixed in the well-established smooth motion manifold. Incorporating an attention mechanism into BRNN achieves better optimization results and higher accuracy because a higher weight value is selectively given to the more important input pose at a specific frame for encoding the input motion when compared to other deep learning frameworks. The results show that our approach efficiently handles various types of motion and noise. We also experiment with different features to find the best feature and believe that our method will be sufficiently desirable to be used in motion capture applications as a post-processing step after capturing human motion.

The biometrics based on resting state electroencephalography (EEG) is better than other EEG-based authentication protocols in terms of usability because it does not require any external stimuli and has a relatively short authentication time. Most of previous resting state EEG-based authentication systems have used a relatively long EEG data (e.g., > 1 min) measured once, and they were segmented to create many trials (e.g., > 100). In this case, however, it is difficult to reflect real-authentication situations in which a user repetitively uses an authentication system in different time points. Therefore, we propose to use single trials repetitively measured for short time (10 s). In the experiment, resting state EEGs were measured while fifteen subjects opened and closed their eyes 30 times for 10 s each. The measured EEG data were divided into three conditions, which are eyes open (EO), eyes closed (EC), and difference between EC and EO (Diff). We extracted power spectral density (PSD) ranging from 3 to 20 Hz as features for classification, with which a binary classification based on a 5×5-fold cross-validation was performed for each subject using linear discriminant analysis (LDA). The mean authentication accuracies of EC, EO, and Diff were 97.05 ± 5.4, 92.5 ± 8.2, and 85.3 ± 7.0 %, respectively, demonstrating the feasibility of single-trial-based EEG authentication. EC could be an optimal condition for developing a resting-state EEG authentication system based on single trial.

In this paper, we present a novel motion retargeting system by using the deep autoencoder combining the Deep Convolution Inverse Graphics Network (DC-IGN) ([Kulkarni et al. 2015]) and the U-Net ([Long et al. 2015]) to produce high-quality human motion. The retargeted motion is fully-automatically and naturally generated from the given input motion and bone length ratios. To validate the proposed motion retargeting system, we conduct several experiments and achieve more accuracy and less computational burden when compared with the conventional motion retargeting approach and other neural network architectures.