No abstract

Skeleton-based human action recognition has attracted much attention in the field of computer vision. Most of the previous studies are based on fixed skeleton graphs so that only the local physical dependencies among joints can be captured, resulting in the omission of implicit joint correlations. In addition, under different views, the content of the same action is very different. In some views, keypoints will be blocked, which will cause recognition errors. In this paper, an action recognition method based on distance vector and multihigh view adaptive network (DV-MHNet) is proposed to address this challenging task. Among the mentioned techniques, the multihigh (MH) view adaptive networks are constructed to automatically determine the best observation view at different heights, obtain complete keypoints information of the current frame image, and enhance the robustness and generalization of the model to recognize actions at different heights. Then, the distance vector (DV) mechanism is introduced on this basis to establish the relative distance and relative orientation between different keypoints in the same frame and the same keypoints in different frame to obtain the global potential relationship of each keypoint, and finally by constructing the spatial temporal graph convolutional network to take into account the information in space and time, the characteristics of the action are learned. This paper has done the ablation study with traditional spatial temporal graph convolutional networks and with or without multihigh view adaptive networks, which reasonably proves the effectiveness of the model. The model is evaluated on two widely used action recognition benchmarks (NTU-RGB + D and PKU-MMD). Our method achieves better performance on both datasets.

Human pose estimation is still a challenging task in computer vision, especially in the case of camera view transformation, joints occlusions and overlapping, the task will be of ever-increasing difficulty to achieve success. Most existing methods pass the input through a network, which typically consists of high-to-low resolution sub-networks that are connected in series. Still, during the up-sampling process, the spatial relationships and details might be lost. This paper designs a parallel atrous convolutional network with body structure constraints (PAC-BCNet) to address the problem. Among the mentioned techniques, the parallel atrous convolution (PAC) is constructed to deal with scale changes by connecting multiple different atrous convolution sub-networks in parallel. And it is used to extract features from different scales without reducing the resolution. Besides, the body structure constraints (BC), which enhance the correlation between each keypoint, are constructed to obtain better spatial relationships of the body by designing keypoints constraints sets and improving the loss function. In this work, a comparative experiment of the serial atrous convolution, the parallel atrous convolution, the ablation study with and without body structure constraints are conducted, which reasonably proves the effectiveness of the approach. The model is evaluated on two widely used human pose estimation benchmarks (MPII and LSP). The method achieves better performance on both datasets.