Human pose estimation has attracted enormous interest in the field of human action recognition. When the human pose is complex (such as pose distortion, pose reversal, etc.) or there is background interference (multi-target, shadow, etc.), the keypoints obtained by existing methods of human pose estimation often have incorrect positioning, category, and connection. This paper proposes a novel human pose estimation network KACNet via the keypoint association constraints. The Channel-1 of KACNet is constrained by the distance loss function to obtain the position of keypoints, and the Channel-2 of KACNet is constrained by the association loss function to obtain the relationship of keypoints. Then, the position and relationship of keypoints are fused by the weighted loss function to obtain the keypoints with accurate location, classification, and connection. Experiments on a large number of public datasets and Internet data show that our method can effectively suppress background interference to improve the accuracy of complex human pose estimation. Compared with state-of-the-art human pose estimation methods, the proposed methods can accurately locate, classify, and connect the human body keypoints robustly.

No abstract

No abstract

No abstract

Objective.
The gait phase and joint angle are two essential and complementary components of kinematics during normal walking, whose accurate prediction is critical for lower-limb rehabilitation, such as controlling the exoskeleton robots. Multi-modal signals have been used to promote the prediction performance of the gait phase or joint angle separately, but it is still few reports to examine how these signals can be used to predict both simultaneously. 
Approach.
To address this problem, we propose a new method named transferable multi-modal fusion (TMMF) to perform a continuous prediction of knee angles and corresponding gait phases by fusing multi-modal signals. Specifically, TMMF consists of a multi-modal signal fusion block, a time series feature extractor, a regressor, and a classifier. The multi-modal signal fusion block leverages the Maximum Mean Discrepancy to reduce the distribution discrepancy across different modals in the latent space, achieving the goal of transferable multi-modal fusion. Subsequently, by using the long short-term memory-based network, we obtain the feature representation from time series data to predict the knee angles and gait phases simultaneously. To validate our proposal, we design an experimental paradigm with random walking and resting to collect data containing multi-modal biomedical signals from electromyography, gyroscopes, and virtual reality.
Main results.
Comprehensive experiments on our constructed dataset demonstrate the effectiveness of the proposed method. TMMF achieves a root mean square error of 0.090±0.022 s in knee angle prediction and a precision of 83.7±7.7\% in gait phase prediction.
Significance.
We demonstrate the feasibility and validity of using TMMF to predict lower-limb kinematics continuously from multi-modal biomedical signals. This proposed method represents application potential in predicting the motor intent of patients with different pathologies.