Abstract Human activity recognition has a wide range of applications in various fields, such as video surveillance, virtual reality and human–computer intelligent interaction. It has emerged as a ...significant research area in computer vision. GCN (Graph Convolutional networks) have recently been widely used in these fields and have made great performance. However, there are still some challenges including over-smoothing problem caused by stack graph convolutions and deficient semantics correlation to capture the large movements between time sequences. Vision Transformer (ViT) is utilized in many 2D and 3D image fields and has surprised results. In our work, we propose a novel human activity recognition method based on ViT (HAR-ViT). We integrate enhanced AGCL (eAGCL) in 2s-AGCN to ViT to make it process spatio-temporal data (3D skeleton) and make full use of spatial features. The position encoder module orders the non-sequenced information while the transformer encoder efficiently compresses sequence data features to enhance calculation speed. Human activity recognition is accomplished through multi-layer perceptron (MLP) classifier. Experimental results demonstrate that the proposed method achieves SOTA performance on three extensively used datasets, NTU RGB+D 60, NTU RGB+D 120 and Kinetics-Skeleton 400.
Deformable attention only focuses on a small group of key sample-points around the reference point and make itself be able to capture dynamically the local features of input feature map without ...considering the size of the feature map. Its introduction into point cloud registration will be quicker and easier to extract local geometric features from point cloud than attention. Therefore, we propose a point cloud registration method based on Spatial Deformable Transformer (SDT). SDT consists of a deformable self-attention module and a cross-attention module where the deformable self-attention module is used to enhance local geometric feature representation and the cross-attention module is employed to enhance feature discriminative capability of spatial correspondences. The experimental results show that compared to state-of-the-art registration methods, SDT has a better matching recall, inlier ratio, and registration recall on 3DMatch and 3DLoMatch scene, and has a better generalization ability and time efficiency on ModelNet40 and ModelLoNet40 scene.
Current brain network studies based on persistent homology mainly focus on the spatial evolution over multiple spatial scales, and there is little research on the evolution of a spatiotemporal brain ...network of Alzheimer's disease (AD). This paper proposed a persistent homology-based method by combining multiple temporal windows and spatial scales to study the spatiotemporal evolution of brain functional networks. Specifically, a time-sliding window method was performed to establish a spatiotemporal network, and the persistent homology-based features of such a network were obtained. We evaluated our proposed method using the resting-state functional MRI (rs-fMRI) data set from Alzheimer's Disease Neuroimaging Initiative (ADNI) with 31 patients with AD and 37 normal controls (NCs). In the statistical analysis experiment, most network properties showed a better statistical power in spatiotemporal networks than in spatial networks. Moreover, compared to the standard graph theory properties in spatiotemporal networks, the persistent homology-based features detected more significant differences between the groups. In the clustering experiment, the brain networks on the sliding windows of all subjects were clustered into two highly structured connection states. Compared to the NC group, the AD group showed a longer residence time and a higher window ratio in a weak connection state, which may be because patients with AD have not established a firm connection. In summary, we constructed a spatiotemporal brain network containing more detailed information, and the dynamic spatiotemporal brain network analysis method based on persistent homology provides stronger adaptability and robustness in revealing the abnormalities of the functional organization of patients with AD.
Despite the severe social burden caused by Alzheimer's disease (AD), no drug than can change the disease progression has been identified yet. The structural brain network research provides an ...opportunity to understand physiological deterioration caused by AD and its precursor, mild cognitive impairment (MCI). Recently, persistent homology has been used to study brain network dynamics and characterize the global network organization. However, it is unclear how these parameters reflect changes in structural brain networks of patients with AD or MCI. In this study, our previously proposed persistent features and various traditional graph-theoretical measures are used to quantify the topological property of white matter (WM) network in 150 subjects with diffusion tensor imaging (DTI). We found significant differences in these measures among AD, MCI, and normal controls (NC) under different brain parcellation schemes. The decreased network integration and increased network segregation are presented in AD and MCI. Moreover, the persistent homology-based measures demonstrated stronger statistical capability and robustness than traditional graph-theoretic measures, suggesting that they represent a more sensitive approach to detect altered brain structures and to better understand AD symptomology at the network level. These findings contribute to an increased understanding of structural connectome in AD and provide a novel approach to potentially track the progression of AD.
Recent research of persistent homology in algebraic topology has shown that the altered network organization of human brain provides a promising indicator of many neuropsychiatric disorders and ...neurodegenerative diseases. However, the current slope-based approach may not accurately characterize changes of persistent features over graph filtration because such curves are not strictly linear. Moreover, our previous integrated persistent feature (IPF) works well on an rs-fMRI cohort while it has not yet been studied on metabolic brain networks. To address these issues, we propose a novel univariate network measurement, kernel-based IPF (KBI), based on the prior IPF, to quantify the difference between IPF curves. In our experiments, we apply the KBI index to study fluorodeoxyglucose positron emission tomography (FDG-PET) imaging data from 140 subjects with Alzheimer's disease (AD), 280 subjects with mild cognitive impairment (MCI), and 280 healthy normal controls (NC). The results show the disruption of network integration in the progress of AD. Compared to previous persistent homology-based measures, as well as other standard graph-based measures that characterize small-world organization and modular structure, our proposed network index KBI possesses more significant group difference and better classification performance, suggesting that it may be used as an effective preclinical AD imaging biomarker.
Aiming to problems in the pairwise registration of point clouds, such as keypoints are difficult to describe accurately, corresponding points are difficult to match accurately and convergence speed ...is slow due to uncertainty of initial transformation matrix, we propose a novel feature descriptor based on ratio of rotational volume to describe effectively keypoints, and on the basis of the feature descriptor, we proposed an improved coarse-to-fine registration pipeline of point clouds, in which we use coarse registration to obtain a good initial transformation matrix and then use fine registration based on a modified ICP algorithm to obtain an accurate transformation matrix. Experimental results show that our proposed feature descriptor has a good robustness to rotation, noise, scale and varying mesh resolution, less storage space and faster running speed than PFH, FPFH, SHOT and RoPS descriptors, and our improved pairwise registration pipeline is very effective to solve the problems in the pairwise registration of point clouds.
Current researches on default mode network (DMN) in normal elderly have mainly focused on finding some dysfunctional areas with decreased or increased connectivity. The global network dynamics of ...apolipoprotein E (APOE) e4 allele group is rarely studied. In our previous brain network study, we have demonstrated the advantage of persistent homology. It can distinguish robust and noisy topological features over multiscale nested networks, and the derived properties are more stable. In this study, for the first time we applied persistent homology to analyze APOE-related effects on whole-brain functional network. In our experiments, the risk allele group exhibited lower network radius and modularity in whole brain DMN based on graph theory, suggesting the abnormal organization structure. Moreover, two suggested measures from persistent homology detected significant differences between groups within the left hemisphere and in the whole brain in two datasets. They were more statistically sensitive to APOE genotypic differences than standard graph-based measures. In summary, we provide evidence that the e4 genotype leads to distinct DMN functional alterations in the early phases of Alzheimer's disease using persistent homology approach. Our study offers a novel insight to explore potential biomarkers in healthy elderly populations carrying APOE e4 allele.
Web-based search query data have been recognized as valuable data sources for discovering new influenza epidemics. However, selecting search and query keywords and adopting prediction methods pose ...key challenges to improving the effectiveness of influenza prediction. In this study, web search data were analyzed and excavated using big data and machine learning methods. The flu prediction model for the southern region of China, considering the impact of influenza transmission across regions and based on various keywords and historical influenza-like illness percentage (ILI%) data, was built (models 1–4) to verify the factors affecting the spread of the flu. To improve the accuracy of the influenza trend prediction, a support vector regression method based on an improved particle swarm optimization algorithm was proposed (IPSO-SVR), which was applied to the influenza prediction model to forecast ILI% in southern China. By comparing and analyzing the prediction results of each model, model 4, using the IPSO-SVR algorithm, exhibited higher prediction precision and more effective results, with its prediction indexes including the mean square error (MSE), root mean square error (RMSE) and mean absolute error (MAE) being 0.0596, 0.2441 and 0.1884, respectively. The experimental results show that the prediction precision significantly increased when the IPSO-SVR method was applied to the constructed ILI% model. A new theoretical basis and implementation strategy were provided for achieving more accurate influenza prevention and control in southern China.
Due to the huge difference in the representation of sketches and 3D models, sketch-based 3D model retrieval is a challenging problem in the areas of graphics and computer vision. Some ...state-of-the-art approaches usually extract features from 2D sketches and produce multiple projection views of 3D models, and then select one view of 3D models to match sketch. It's hard to find "the best view" and views from different perspectives of a 3D model may be completely different. Other methods apply learning features to retrieve 3D models based on 2D sketch. However, sketches are abstract images and are usually drawn subjectively. It is difficult to be learned accurately. To address these problems, we propose cross-domain correspondence method for sketch-based 3D model retrieval based on manifold ranking. Specifically, we first extract learning features of sketches and 3D models by two-parts CNN structures. Subsequently, we generate cross-domain undirected graphs using learning features and semantic labels to create correspondence between sketches and 3D models. Finally, the retrieval results are computed by manifold ranking. Experimental results on SHREC 13 and SHREC 14 datasets show the superior performance in all 7 standard metrics, compared to the state-of-the-art approaches.
In the fields of computer graphics and computer vision, a great amount of research and analysis has been conducted on expression-carrying face models. How to construct more realistic and effective 3D ...face models has become an immense challenge to researchers. This paper proposes a parametric 3D face model editing algorithm based on existing 3D face models. The algorithm takes a number of existing 3D face models with different expressions as input, edits the models through mostly through model deformation and interpolation, and generates a new 3D face model. In particular, the face model editing process begins with selecting multiple face models with different expressions as input. Second, with one of the selected models as the source model and others as the target models, the source model and all target models are registered one by one; meanwhile, the vertex correspondence between the registered models is established. Third, the selected 3D models are parameterized to a planar disc through quasi-conformal mapping. Fourth, relying on the vertex correspondence, a set of corresponding control points between different models are established. The model is then deformed and interpolated under the guidance of the control points and by using the quasi-conformal iteration method, which produces the 2D face models with transitional expressions between the source model and the target models. Finally, the 2D models are restored to the corresponding 3D face models using the model restoration algorithm. Additionally, this paper proposes to use the Beltrami coefficient to guide the quasi-conformal iteration in performing the mapping between two planes. This coefficient then serves as a measure to evaluate the similarity between the edited model and the original one. The proposed algorithm has been evaluated through extensive experiments. The results suggest that compared with existing editing methods, the proposed method is more effective and efficient in constructing various 3D face models.