Using Handheld Depth Sensor for 3D Model Reconstruction and Augmented Reality Application

• Main Authors: HUNG, CHIEN-YU, 洪千喻
• Other Authors: Hwang, Jin-Tsong
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/m2fvhr

碩士 === 國立臺北大學 === 不動產與城鄉環境學系 === 107 === In recent years, 3D modeling technology has developed rapidly, and the application has become increasingly diverse, such as virtual reality and augment reality. While the acquisition of 3d information, such as the passive sensing of photogrammetry and multi-angle image modeling based on computer vision, and the active sensing of LiDAR etc., have their own advantages and disadvantages, like the modeling of passive sensing is difficult in the area with insufficient feature points, and the high-precision active sensing instrument is expensive. This study uses Occipital Structure Sensor as the test instrument, which is an active infrared ray range finder and can be attached to the mobile device. The product has corresponding development software, which is both convenient and economical. However, it is necessary to understand the 3D model precision constructed by Structure Sensor, in order to understand the applicable fields, limitations and problems. In this study, the volume value and the distance of feature points were calculated as the 3D accuracy evaluation method, and a 3D point cloud was obtained by using Skannect software. After reconstruction of the point cloud model, the volume and the distance of feature points were calculated using The Maximum Likelihood Estimation Sample Consensus (MLESAC) to extract the model with geometric significance (square body, sphere and cylinder) . The experiment shows that the 3D model volume error of the Structure Sensor construction is in the range of 1.69% ~ 5.30%, and the feature point distance error is 1.09% ~ 2.64%.This method is convenient to obtain 3D information, and also provides another 3D modeling choice for texture-less object in the future. In terms of the application of the 3D model, this study established the augmented reality application of teaching level’s operational process through Google ARCore and game development engine Unity. Through this research process, the level model was built and the point-select sliding function of each part of the level was designed to provide simulation operation experience and improve the application of 3D modeling.