<p>Three-dimensional (3D) reconstruction is a technology that extracts 3D information—such as key points that can identify the corners and brightness of an object—from photographs taken by a camera or data scanned by a computer. Conventional 3D reconstruction methods use either still images or videos. Although methods using still images are more commonly used for 3D reconstruction, they require a large number of still images to perform highly accurate reconstruction. In this study, we explore a technique for generating highly accurate 3D models using reconstruction methods using videos. In this paper, we propose a method for high-precision 3D reconstruction from video sequences obtained from low-quality cameras, such as smartphone cameras, by calculating the sharpness of frames using the Canny filter and selecting frames with high sharpness through peak estimation. We also investigate the impact of parameters in the Canny method and peak estimation on reconstruction quality. As a result, we found that highly accurate 3D reconstruction is achievable using the parameter values identified as optimal in this paper.</p>