摘要: | 在當今的數位時代,三維掃描和重建技術在自動駕駛、精密工程和文化遺產保存等多個領域中變得至關重要,允許我們準確地創建物理對象的數位模型以進行分析和虛擬互動。然而,這些先進技術在實施過程中經常會受到各種雜訊的干擾,這些雜訊可能來自於影像特徵識別錯誤、環境條件或物體表面的複雜性,從而在3D重建過程中造成物體特徵的缺失,對物體特徵的重建和描述構成重大挑戰。 本研究提出一種基於運動恢復結構與熵計算相結合的創新三維點雲重建方法。通過在圖像處理階段引入熵計算,此方法有效地提升對極幾何的豐富度和點雲數量。此外,改進的密度熵束調整使點雲能夠更佳的豐富,展現更細緻紋理的三維點雲模型。最終,將所得的點雲模型整合到虛幻引擎中,使其具備即時更新和高度逼真的視覺效果,展示本研究在3D重建領域的潛在應用和創新價值。 實驗結果表明,此方法在點雲密度和細節豐富性方面顯著優於現有的SfM方法。儘管計算時間較長,但其生成的點雲模型在密度和細節上具有明顯的優勢,這對於需要詳細三維重建的應用尤其有利。通過將此方法應用於工業相機捕捉的圖像中,成功創建數位孿生模型,並將其導入虛幻引擎中,實現即時監控。不僅展示所提出方法的實際應用潛力,還強調其在數位孿生和虛擬現實領域的創新價值。 ;In today’s digital era, 3D scanning and reconstruction technologies have become pivotal in various fields such as autonomous driving, precision engineering, and cultural heritage preservation, allowing us to accurately create digital model of physical objects for analysis and virtual interactions. However, these advanced technologies often encounter obstruction from various types of noise during implementation, which from misidentification of image features, environmental conditions, or the complexity of object surfaces, thereby causing loss of object features in the 3D reconstruction process and posing significant challenges to the reconstruction and description of object features. Therefore, developing efficient methods for noise identification and reduction becomes particularly important to enhance the accuracy and reliability of 3D scanning and reconstruction technologies. This study proposes an innovative 3D point cloud reconstruction method based on Structure from Motion (SfM) and entropy calculation. By introducing entropy calculation in the image processing phase, this method effectively enhances the richness of epipolar geometry and point clouds numbers. Additionally, the improved density entropy Bundle Adjustment allows point clouds to be more detailed, presenting a more refined texture in the 3D point cloud model. Finally, the resulting point cloud model is integrated into the Unreal Engine, enabling real-time updates and highly realistic visual effects, displaying the potential applications and innovative value of this research in the field of 3D reconstruction. Experimental results show that this method significantly outperforms existing SfM methods in terms of point cloud density and detail richness. Although the computation time is longer, the generated point cloud model has clear advantages in density and detail, which is especially beneficial for applications requiring detailed 3D reconstruction. By applying this method to images captured by industrial cameras, it successfully creates a digital twin model and integrates it into the Unreal Engine for real-time monitoring. This not only demonstrates the practical application potential of the proposed method but also highlights its innovative value in the fields of digital twins and virtual reality. |