Abstract:The quality of collision detection algorithm directly affects the performance of the whole simulation system. To address the low efficiency and low accuracy in detecting the collisions of flexible cloths in virtual environments, this paper proposes an oriented bounding box (OBB) algorithm with a simplified model, tree structure for a root-node double bounding box, and continuous collision detection algorithm incorporating an OpenNN-based neural network optimization. First, for objects interacting with the cloths with more complex modeling, the model is simplified with a surface simplification algorithm based on the quadric error metrics, and the simplified model is used to construct an OBB. Second, a bounding box technique commonly used for collision detection is improved, and a root-node double bounding box algorithm is proposed to reduce the construction time for the bounding box. Finally, neural networks are used to optimize the continuous collision detection algorithm, as neural networks can efficiently process large amounts of data and remove disjoint collision pairs. An experiment shows that the construction of an OBB using the simplified model is almost identical to that of the original model, but the taken to construct the OBB is reduced by a factor of approximately 2.7. For the same cloth, it takes 5.51%–11.32% less time to run the root-node double bounding box algorithm than the traditional-hybrid bounding box algorithm. With an average removal rate nearly identical to that of the traditional filtering method, the elapsed time is reduced by 7%–11% by using the continuous collision detection algorithm based on an OpenNN neural network optimization. The simulation results are realistic and in line with the requirements for real-time cloth simulations.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to detect collisions between flexible cloths efficiently and accurately in a virtual environment. Specifically, the author points out that the existing collision detection algorithms have the problems of low efficiency and poor accuracy when dealing with complex - modeled flexible cloths. To solve these problems, the paper proposes an improved method based on the Oriented Bounding Box (OBB) algorithm, including model simplification, the double - layer bounding box technique for the root node of the tree structure, and the continuous collision detection algorithm optimized in combination with the OpenNN neural network. ### Main Contributions 1. **Model Simplification**: - Use the surface simplification algorithm based on quadratic error metric to simplify complex - modeled objects, thereby reducing the time for constructing OBB. - The simplified model can maintain the main features of the original model while significantly reducing the amount of calculation. 2. **Root Node Double - Layer Bounding Box Algorithm**: - Propose a new root node double - layer bounding box algorithm, which reduces the construction time of bounding boxes by using double - layer bounding boxes at the root node. - This method combines the advantages of bounding spheres and AABBs and improves the compactness of bounding boxes. 3. **Continuous Collision Detection Algorithm Optimized Based on OpenNN Neural Network**: - Utilize the ability of neural networks to efficiently process large amounts of data to optimize the continuous collision detection algorithm. - Through training the neural network to learn algebraic non - penetration filtering and spatial linear projection filtering, quickly remove non - intersecting collision pairs. ### Experimental Results - The time for constructing OBB using the simplified model is reduced by approximately 2.7 times. - The root node double - layer bounding box algorithm saves 5.51% to 11.32% of the running time compared to the traditional hybrid bounding box algorithm. - The continuous collision detection algorithm optimized based on OpenNN neural network reduces the running time by 7% to 11% while maintaining a removal rate similar to that of the traditional filtering methods. These improvements make the collision detection in cloth simulation more efficient and accurate, meeting the requirements of real - time cloth simulation.

Deep Neural Network-Based Cloth Collision Detection Algorithm

Collision detection for clothed human animation

D-Cloth: Skinning-based Cloth Dynamic Prediction with a Three-stage Network

Avenues of Dynamic Cloth Modeling

A Coherence-Based Collision Detection Method For Dressed Human Simulation

A Novel Collision Detection Algorithm Based on Grids Enveloping for Industrial Robot Simulations

Cloth simulation using multilevel meshes

I-cloth: incremental collision handling for GPU-based interactive cloth simulation.

I-cloth

A Unified Cloth Untangling Framework Through Discrete Collision Detection

ICCD: Interactive Continuous Collision Detection Between Deformable Models Using Connectivity-Based Culling

Deformation Similarity Clustering Based Collision Detection In Clothing Simulation

Hierarchical Cloth Simulation using Deep Neural Networks

Dynamic Cloth Motion Simulation with a Novel Fast Collision Detection

Fast and accurate continuous collision detection between rigid bodies

SENC: Handling Self-collision in Neural Cloth Simulation

Neural Collision Detection for Deformable Objects

Nonlinear Cloth Simulation with Isogeometric Analysis

A Neural-Network-Based Approach for Loose-Fitting Clothing

Optimized Space Decomposition Algorithm for Parallel Continuous Collision Detection

Cloth simulation based on local adaptive subdivision and merging