Current reconstruction systems often face the challenge of drifting when reconstructing complex scenes. Recent 3D(three-dimensional) reconstruction systems have shown convincing results, but still suffer from the following problems: (1) When the current vision-based 3D reconstruction system uses a single camera , the small angle of view of the camera is likely to cause the reconstructed 3D model to be incomplete. (2) Some image frames have fewer image feature points and image blurring, which leads to a larger deviation of the estimated camera pose value. (3) The current mainstream line feature 3D reconstruction system causes linearization and limits the update efficiency due to the adoption of the filter frame. In order to solve the above problems, this paper proposes a highly robust automatic 3D reconstruction system based on integrated optimization by point line features. Firstly, a multi-depth camera collaborative scanning method is developed to obtain a relatively complete 3D model. Secondly, a more accurate camera pose initial value can be obtained in advance without the position estimation. Thirdly, a comprehensive optimization method based on point line feature is used, which can improve the accuracy of camera pose and the consistency and accuracy of map construction. Many experiments show that the system can solve the problems of small viewing angle, blurred image and low modeling efficiency. The proposed system can be applied to 3D reconstruction of various complex large scenes. The obtained high-precision 3D model can be widely applied in the fields of human–computer interaction, virtual reality, etc.

当前的重建系统在重建复杂场景时经常面临漂移的挑战。最近的3D（三维）重建系统已显示出令人信服的结果，但仍存在以下问题：（1）当当前基于视觉的3D重建系统使用单个摄像机时，摄像机的小视角可能会导致重建的3D模型不完整。（2）一些图像帧具有较少的图像特征点和图像模糊，这导致估计的相机姿态值的较大偏差。（3）当前的主流线特征3D重建系统由于采用了滤镜框架而导致线性化并限制了更新效率。为了解决上述问题，本文提出了一种基于点线特征的集成优化的高度健壮的自动3D重建系统。首先，开发了一种多深度相机协同扫描方法以获得相对完整的3D模型。其次，可以在不进行位置估计的情况下预先获得更准确的相机姿势初始值。第三，采用基于点线特征的综合优化方法，可以提高摄像机姿态的精度以及地图构造的一致性和准确性。许多实验表明，该系统可以解决视角小，图像模糊，建模效率低的问题。所提出的系统可以应用于各种复杂的大型场景的3D重建。所获得的高精度3D模型可以广泛应用于人机交互领域，