振动
计算机视觉
帧(网络)
计算机科学
帧速率
人工智能
声学
物理
电信
作者
Wenxiang Qin,Feiyue Wang,Shaopeng Hu,Kohei Shimasaki,Idaku Ishii
标识
DOI:10.1109/tim.2025.3593568
摘要
Vibration measurement is crucial for assessing the durability and reliability of structures and products. While 3D digital image correlation (DIC) methods based on stereo image analysis are widely used to visualize 3D vibration distributions, their performance significantly degrades on weakly textured surfaces due to its reliance on substantial textures or artificial patterns for stereo correspondence. This dependency limits robustness, accuracy, and processing efficiency. To overcome these limitations, we propose a model-based 3D-DIC method that enables robust and accurate measurement of 3D vibration distributions with minimal dependency on surface patterns. Using semi-rigid 3D shape priors, our method projects model points onto stereo images and reconstructs 3D vibrations through the triangulation of 2D-DIC displacements. The projection strategy eliminates the need for complex stereo correspondence, thereby improving the processing efficiency. A cuboid experiment with 0–4 mm depth movements demonstrated an accuracy of 34 μm under low-contrast conditions, outperforming conventional 3D-DIC methods without shape priors. Dynamic measurement performance was assessed using a repetitive-texture metal plate vibrating at various frequencies, where a 450 Hz vibration with a peak amplitude below 20 μm was detected with an error under 3 μm. Our method precisely visualizes 3D vibration distributions at 40 Hz and 100 Hz on a low-texture PC stereo speaker. These results demonstrate that our method provides a promising solution for the accurate, robust, and efficient measurement and analysis of 3D vibrations in weakly patterned objects.
科研通智能强力驱动
Strongly Powered by AbleSci AI