Robust 3D Reconstruction Through Noise Reduction of Ultra-Fast Images

3D reconstruction from multiple view images has been studied extensively in computer vision tasks. In order to increase the accuracy of the 3D reconstruction, it is important to secure the number of image frames and to find feature points and match accurate feature points by minimizing the influence of noise from each image. When we acquired images from high-speed camera, it is possible to analyze phenomena and object movements that are difficult to see with the naked eye. However, when using a high-speed camera, problems such as increased data amount, light amount, focus, and noise occur due to an increase in resolution and shutter speed. In this paper, we propose a preprocessing method for feature point tracking and matching for robust 3D reconstruction in high-speed images. The experimental results confirm the validity compared with 3D reconstruction output from the original image and preprocessed image.