为了克服三维重建高度依赖标定板, 满足3D打印模型的工业需求, 提出基于图像自标定的高效3D打印模型生成方法, 无需借助标定板计算相机参数, 直接使用单相机采集序列图像进行三维重建。为了克服基于自标定方法易受图像质量和特征点匹配精确度的影响, 根据人机交互与自适应分割算法相结合的方法去除原始图像背景及过滤噪声, 使图像感兴趣区域特征更为明显, 采用快速稳定特征算法提取序列图像中特征点并根据特征点的匹配度进行精确的特征点匹配, 再使用匹配信息自标定求解得到相机模型参数, 最后根据相机模型以及特征点信息完成三维目标的稠密重建。实验结果表明, 自标定及重建方法对大小各异, 表面材质不同的目标均可实现重建。

In order to improve the performance of three-dimensional reconstruction and overcome the highly dependent relationship on calibration board, an efficient generation 3D printing model method was proposed. Without using the calibration board to calculate the camera parameters in this method, the image captured by single camera can be used to generate 3D model. However, this self-calibration method is deeply influenced by the image quality and point matching accuracy, which limits the 3D printing model generation with high efficiency. In order to overcome these effects, firstly the background area and restrain noise are removed by interactive and graph partition to enhance the image’s region-of-tnterest(ROI).Second feature point is extracted from sequences pictures by improved speed-up robust features (SURF) algorithm and is matched based on its’ matching factor, which shows faster than the previous program. Then camera model parameters are calculated by self-calibration matching information. Finally, dense 3D object is reconstructed by combining camera model and feature point matching information. A series of experiments show the proposed method is characterized by effectiveness, convenience and wide application.