Using BlazePose on Spatial Temporal Graph Convolutional Networks for Action Recognition

The ever-growing available visual data(i.e.,uploaded videos and pictures by internet users)has attracted the research community’s attention in the computer vision field.Therefore,finding efficient solutions to extract knowledge from these sources is imperative.Recently,the BlazePose system has been released for skeleton extraction from images oriented to mobile devices.With this skeleton graph representation in place,a Spatial-Temporal Graph Convolutional Network can be implemented to predict the action.We hypothesize that just by changing the skeleton input data for a different set of joints that offers more information about the action of interest,it is possible to increase the performance of the Spatial-Temporal Graph Convolutional Network for HAR tasks.Hence,in this study,we present the first implementation of the BlazePose skeleton topology upon this architecture for action recognition.Moreover,we propose the Enhanced-BlazePose topology that can achieve better results than its predecessor.Additionally,we propose different skeleton detection thresholds that can improve the accuracy performance even further.We reached a top-1 accuracy performance of 40.1%on the Kinetics dataset.For the NTU-RGB+D dataset,we achieved 87.59%and 92.1%accuracy for Cross-Subject and Cross-View evaluation criteria,respectively.

S^(2)ANet:Combining local spectral and spatial point grouping for point cloud processing

Background Despite the recent progress in 3D point cloud processing using deep convolutional neural networks,the inability to extract local features remains a challenging problem.In addition,existing methods consider only the spatial domain in the feature extraction process.Methods In this paper,we propose a spectral and spatial aggregation convolutional network(S^(2)ANet),which combines spectral and spatial features for point cloud processing.First,we calculate the local frequency of the point cloud in the spectral domain.Then,we use the local frequency to group points and provide a spectral aggregation convolution module to extract the features of the points grouped by the local frequency.We simultaneously extract the local features in the spatial domain to supplement the final features.Results S^(2)ANet was applied in several point cloud analysis tasks;it achieved stateof-the-art classification accuracies of 93.8%,88.0%,and 83.1%on the ModelNet40,ShapeNetCore,and ScanObjectNN datasets,respectively.For indoor scene segmentation,training and testing were performed on the S3DIS dataset,and the mean intersection over union was 62.4%.Conclusions The proposed S^(2)ANet can effectively capture the local geometric information of point clouds,thereby improving accuracy on various tasks.

Skeleton Split Strategies for Spatial Temporal Graph Convolution Networks

Action recognition has been recognized as an activity in which individuals’behaviour can be observed.Assembling profiles of regular activities such as activities of daily living can support identifying trends in the data during critical events.A skeleton representation of the human body has been proven to be effective for this task.The skeletons are presented in graphs form-like.However,the topology of a graph is not structured like Euclideanbased data.Therefore,a new set of methods to perform the convolution operation upon the skeleton graph is proposed.Our proposal is based on the Spatial Temporal-Graph Convolutional Network(ST-GCN)framework.In this study,we proposed an improved set of label mapping methods for the ST-GCN framework.We introduce three split techniques(full distance split,connection split,and index split)as an alternative approach for the convolution operation.The experiments presented in this study have been trained using two benchmark datasets:NTU-RGB+D and Kinetics to evaluate the performance.Our results indicate that our split techniques outperform the previous partition strategies and aremore stable during training without using the edge importance weighting additional training parameter.Therefore,our proposal can provide a more realistic solution for real-time applications centred on daily living recognition systems activities for indoor environments.

Automatic Segmentation Method for Cone-Beam Computed Tomography Image of the Bone Graft Region within Maxillary Sinus Based on the Atrous Spatial Pyramid Convolution Network

Sinus floor elevation with a lateral window approach requires bone graft(BG)to ensure sufficient bone mass,and it is necessary to measure and analyse the BG region for follow-up of postoperative patients.However,the BG region from cone-beam computed tomography(CBCT)images is connected to the margin of the maxillary sinus,and its boundary is blurred.Common segmentation methods are usually performed manually by experienced doctors,and are complicated by challenges such as low efficiency and low precision.In this study,an auto-segmentation approach was applied to the BG region within the maxillary sinus based on an atrous spatial pyramid convolution(ASPC)network.The ASPC module was adopted using residual connections to compose multiple atrous convolutions,which could extract more features on multiple scales.Subsequently,a segmentation network of the BG region with multiple ASPC modules was established,which effectively improved the segmentation performance.Although the training data were insufficient,our networks still achieved good auto-segmentation results,with a dice coefficient(Dice)of 87.13%,an Intersection over Union(Iou)of 78.01%,and a sensitivity of 95.02%.Compared with other methods,our method achieved a better segmentation effect,and effectively reduced the misjudgement of segmentation.Our method can thus be used to implement automatic segmentation of the BG region and improve doctors’work efficiency,which is of great importance for developing preliminary studies on the measurement of postoperative BG within the maxillary sinus.