Akademik Veri Yönetim Sistemi
20th Conference on Computer Science and Intelligence Systems (FedCSIS), Krakow, Polonya, 14 - 17 Eylül 2025, cilt.43, ss.309-314, (Tam Metin Bildiri)
In this study, we present a data-driven method for preserving the length of cubic B´ezier curves under external deformations, specifically in the context of hair strand simulation. By sampling millions of original and displaced control point con- figurations within a 2D bounded grid space, we construct a large dataset of B\'ezier curve pairs and compute the corresponding t\_final parameters required to preserve arc length. A lightweight artificial neural network (ANN) architecture is trained on this dataset to predict t\_final given 16 features representing the coor- dinates of initial and displaced control points. The model achieves a low mean absolute error (MAE) of 0.00091992 on the test set, ensuring high predictive accuracy. Performance evaluations show that the ANN can predict t\_final values for up to 200k hair strands in approximately 8 seconds on a standard laptop, aligning with the average number of strands on a human scalp. This approach replaces computationally expensive numerical length-matching operations with a fast, inference-based framework, allowing for efficient simulation of realistic, deformable hair motion while maintaining individual strand lengths.