基于图像识别的SolidWorks三维重建技术研究

Research on SolidWorks 3D Reconstruction Technology Based on Image Recognition

摘要: 本研究聚焦于基于图像识别的三维重建技术，强调高层理解的扫描图纸作为重要前提，并在机械知识的引导下实现三维模型的重建。图像识别分为基本图素识别和字符识别两个部分，其中基本图素识别采用改进的Hough变换法，字符识别采用基于卷积神经网络的方法，以适应实际图样的特殊情景。实验结果表明，该模型具有显著的识别效果，能够有效兼容干扰图线，识别率高达98.65%。随后，通过工程图的语义知识，建立基本图素之间的拓扑关系和尺寸约束图形链，实现对扫描图纸的二维重建。最终，在机械领域知识的支持下，充分利用图纸识别后的信息和知识推理，实现三维形体的重建。

Abstract: This study focuses on 3D reconstruction techniques based on image recognition, with an emphasis on the high-level understanding of scanned drawings as a crucial prerequisite. The aim is to realize the reconstruction of 3D models under the guidance of mechanical knowledge. The image recognition process is divided into two parts: basic pixel recognition and character recognition. The improved Hough transform method is adopted for basic pixel recognition, while character recognition utilizes a method based on convolutional neural networks to adapt to the specific characteristics of actual drawings. Experimental results demonstrate that the model has a significant recognition effect, effectively handling interference lines with a high recognition rate of 98.65%. Furthermore, by leveraging semantic knowledge from engineering drawings, the topological relationship between basic graphic elements and size constraint graph chains is established to achieve two-dimensional reconstruction of scanned drawings. Finally, with support from mechanical domain knowledge, 3D shape reconstruction is realized through comprehensive information utilization and knowledge reasoning following drawing recognition.