Computer Animation and Virtual Worlds最新文献

In virtual reality (VR) applications, real-time and robust 3D human pose estimation is paramount to enhance user experience, yet existing methodologies often encounter challenges such as high computational burden, occlusion sensitivity, and inadequate adaptation to complex actions. To mitigate these issues, we propose a novel 3D human pose estimation method based on a multi-agent hierarchical biomechanical priors architecture. This method achieves efficient heatmap prediction in the local feature space through a parallel agents architecture, while simultaneously integrating a hierarchical loss function and dynamic context modeling. It incorporates virtual avatars' geometric constraints into network training, thereby enhancing pose plausibility and effectively addressing occlusion and intricate actions. Moreover, it substantially improves cross-frame stability and estimation accuracy in occlusion scenarios through multiview spatiotemporal consistency optimization. Compared to existing methods, the proposed framework provides adaptability to the unique demands of virtual environments with reduced computational cost. We experimentally validate our approach on the widely used Human3.6M and MPI-INF-3DHP datasets, and further demonstrate through ablation experiments that the dynamic occlusion compensation module, which fuses multimodal perception with a spatio-temporal diffusion mechanism, significantly enhances the robustness of pose estimation under occlusion scenarios with virtual costumes.

This study presents a novel tri-manual interaction framework that enables users to control two physical hands via VR controllers and a virtual robotic arm through a hybrid brain-computer interface (32-channel EEG and eye-tracking). The virtual robotic arm, implemented as a 6-DOF industrial manipulator in Unity, is controlled through simplified BCI commands: forward/backward movement along the z-axis based on motor imagery strength, with automatic grasping triggered by sustained attention thresholds. Twenty-five participants completed 30 trials, each following a 60-s protocol with five phases: rest, target presentation, preparation, execution, and feedback. Results demonstrated that the hybrid BCI system achieved superior performance compared to EEG-only control: 8.5% improvement in task success rate, 34.5% increase in positioning accuracy, and 46.2% reduction in cognitive load. The CNN-LSTM architecture achieved 86.3% motor imagery classification accuracy. Learning effects were observed within trials, with performance plateauing after 3.8 ± 0.7 attempts. Time-frequency analysis revealed hierarchical neural coordination mechanisms underlying the tri-manual control. This research validates the feasibility of augmented manipulation in virtual environments, establishing a foundation for advanced human-robot collaboration in animation and VR applications.

Smoke simulation is a crucial element in entertainment applications, such as movies and video games. In particular, the required smoke texture varies depending on the scene, ranging from smooth textures for cigarette smoke to highly irregular ones for explosions. The texture of smoke is primarily influenced by light sources, scattering properties, and turbulence components. Light sources and scattering properties affect the appearance of smoke during rendering, influencing its color, brightness, and density. Turbulence components significantly influence the shape of smoke. While many methods have been proposed for generating turbulence components, they all require manual adjustment of turbulence parameters, which is both time-consuming and labor-intensive. To address this issue, we propose a method for easily generating the desired turbulence by estimating turbulence parameters from images. In our approach, users input images including the desired turbulence components, and optimal parameters representing those components are automatically estimated. This reduces the time and effort required for parameter adjustment, allowing the desired turbulence components to be represented more efficiently.

The expression of fine details such as fluid flowing through narrow pipes or split by thin plates poses a significant challenge in simulations involving complex boundary conditions. As a hybrid method, the material point method (MPM), which is widely used for simulating various materials, combines the advantages of Lagrangian particles and Eulerian grids. To achieve accurate simulations of fluid flow through narrow pipes, high-resolution uniform grid cells are necessary, but this often leads to inefficient simulation performance. In this article, we present an adaptive boundary material point method that facilitates adaptive subdivision within regions of interest and conducts collision detection across grids of varying sizes. Within this framework, particles interact through grids of differing resolutions. To tackle the challenge of unevenly distributed subdivided particles, we propose a surface reconstruction approach grounded in the color distance field (CDF), which accurately defines the relationship between the particles and the reconstructed surface. Furthermore, we incorporate a mesh refinement technique to enrich the detail of the mesh utilized to mark the grids during subdivision. We demonstrate the effectiveness of our approach in simulating various materials and boundary conditions, and contrast it with existing methods, underscoring its distinctive advantages.

Motion capture options at a high rate, for example, kinematics of flying organisms, are limited. Inertial sensors or marker-based tracking are infeasible due to the mounting requirement. The advent of high-speed cameras being made available for commercial use has facilitated the discovery of intriguing high-rate behaviors. However, analyzing rapid kinematics in three dimensions necessitates using two or more high-speed cameras, which are specialized equipment and come at a significant expense. In this study, we utilize a conventional methodology of capturing synchronized multiple views using a solitary camera and a planar mirror. We introduce a user-friendly software package that facilitates the calibration of this economical setup, enabling the three-dimensional (3D) analysis of high-rate maneuvers and automatic tracking over long video sequences. Within ∼15 min, a single high-speed camera is calibrated and can be used to acquire 3D data. We accompany the toolbox with a detailed user guide and video tutorial (https://tinyurl.com/yckm9y6b) for ease of use. We accurately reconstruct dense tracks for multiple high-rate maneuvers, including flying insect species (dragonfly, butterfly, and housefly), spinning coin, and rotating fan blades. We believe that our affordable setup will assist in gathering high-rate natural phenomena, leading to realistic and diverse simulations.

Labanotation is a scientific method for documenting dance movements that has been widely adopted globally. Existing methods for Labanotation action recognition perform poorly in handling complex movements and integrating spatiotemporal information. To address this, we propose a multi-branch spatiotemporal fusion network with attention mechanisms aimed at accurately recognizing Labanotation actions from motion capture data. Initially, we convert motion capture data into three-dimensional coordinates and extract skeleton vector features. Subsequently, we enhance feature representation by extracting temporal difference features and skeleton angle features from the skeleton vectors. These features are processed using gated recurrent units and residual networks to effectively integrate spatiotemporal information. Finally, attention mechanisms are applied in the model to differentiate the importance of different positions in the features. This method effectively models spatiotemporal relationships, thereby improving the accuracy of Labanotation action recognition. We conducted experiments on two segmented motion capture datasets, demonstrating the effectiveness of each module. Compared to existing methods, our approach shows superior performance and strong generalization ability. Given the relative simplicity of upper limb action recognition, our focus primarily lies on lower limb action recognition. Notably, this marks the first application of skeleton angle features in the field of Labanotation action recognition.