| 摘要: | 從最早的針孔相機開始，到現在的單眼相機，人類就想把生活周遭的物品、事件記錄下來。然而相機只能記錄二維的影像，始終不能把真實世界中的三維資訊記錄下來，所以我們需要建立三維模型的工具以獲得真實世界中三維資訊。
三維模型已經廣泛用在各領域，如在自駕車、掃地機器人上用來測量距離與前方障礙物，在電影拍攝時動態捕捉演員的臉部變化以便於後製動畫的處理，在醫療上牙齒和各個器官的建模可用於觀察病人的健康狀況或建立精確的模型，因此，如何把真實物體的三維資訊更精確的記錄下來成了重要的課題。
現在三維建模多種不同的技術中，以結構光為基礎的技術通常會遇到景深不足、視角造成影像形變的問題，本研究同樣以結構光為基礎，希望能夠改善這兩項問題，此外在沒有參考平面的情況下量測，並提升系統的可移動性。為了增加系統的可移動性，本研究結合微型投影機與手機，以微型投影機提供結構光，再以手機拍攝結構光影像。過去在結構光系統中最常使用是三相位位移法取得相位差後再重建影像，此方法容易在大視角的位置造成形變，所以本研究中結合二進制編碼與相位位移法，二進制編碼是要精準定位樣本上的條紋，找出其對應到基準面上的位置後便可清楚得知樣本對應到相機的視角，並使用三角量測修正在此視角下的像差。因為相機與樣本相對視角的計算可以使用先行紀錄的二進制編碼條紋，所以不需要基準面的資訊也可重建三維資訊。結合相位位移法則是為了進一步在得知條紋位置後輔助獲得更細緻的相位資訊，以取得更完整的三維資訊。
本研究在系統建構完成後以平板、球類及人臉面具做為測試樣本，以平板為樣本是為了測試重建絕對高度的精準度，將平板放置在距離基準面20 cm到60 cm的位置內，重建的誤差在0.4 cm以內；以桌球和接觸球作為已知的樣本，測試重建簡單幾何曲面的精準度，以圓形擬合與球形擬合找出半徑，同樣將桌球和接觸球放置在距離基準面20 cm到60 cm的位置內，桌球與接觸球相對高度誤差分別在0.04 cm以內、0.06 cm以內；最後以人臉面具證明可以重建較複雜結構的三維資訊。
;From the earliest pinhole camera to the Monocular camera, humans want to record the objects and events.However, the camera can only record two-dimensional (2D) images, and can never record the three-dimensional(3D) information in the real world. So we need instruments to build 3D models to preliminary 3D information in the real world.
3D models have been widely used in various fields. Self-driving and Robotic vacuum cleaner measure distance and obstacles in front. The dynamic capture of the human face during film shooting, recording the actor′s face changes to facilitate the post processing. Modeling of the teeth and various organs in medical treatment can be used to observe the health of the patient or to establish an accurate model. How to record real items more accurately becomes another topic.
Nowadays, among many different technologies from 3D modeling, the 3D modeling based on structured light usually faces the problems of lacking depth of field and image distortion caused by angle of view. This study is also based on structured light and hoped these two problems can be improved. In addition, the measurement is performed without a reference plane and the system′s mobility is improved. This study combines a micro-projector with a mobile phone to provide structured light with a micro-projector, and then capture a structured-light image with a mobile phone. In the past, the most commonly algorithm used in structured light systems was three step phase shifting. It reconstructs the image after obtaining the phase difference, but this method is easy to cause deformation in the position of a large angle of view. So this study combine binary coding and phase shifting method. Binary coding located the fringes on the sample accurately, find out the position corresponding to the reference plane, and then clearly know the angle of view of the sample corresponding to the camera, and use triangulation to correct the aberrations at this angle of view. Because the relative angle of view between the camera and the sample can be calculated using the previously recorded binary coded fringes, three-dimensional information can be reconstructed without the need for information on the reference plane. Finally, the combination of the phase shifting method is to further obtain more detailed phase information after obtaining the position of the fringe to obtain more complete three-dimensional information.
In this study, tablets, balls and face masks are used as samples after the system architecture is completed. The plate is used as a sample to test the accuracy of the absolute height of the reconstruction. Placing the plate within 20 cm to 40 cm from the reference plane, the reconstruction error is within 0.5 cm. Using table tennis and contact balls as known samples, test the accuracy of reconstructing simple geometric surfaces. Find the radius with circular fitting and spherical fitting, and also place the table tennis and contact ball within 20 cm to 40 cm from the reference plane. The errors of table tennis and contact ball fits within 0.12 cm and 0.1 cm. Eventually, the face mask proves that it can reconstruct three-dimensional information of more complex structures. |
