基于激光与单目视觉的室内场景三维重建

3D Indoor Reconstruction Based on Laser Scanner and Monocular Camera

利用云台和二维激光测距仪构建起三维激光扫描仪,对室内环境下采集到的激光深度信息与二维图像进行像素级融合,建立彩色三维点云模型。针对该模型对场景的视觉表现力较差问题,结合随机抽样一致性原理与迭代最近点算法从激光点云数据中提取场景的平面框架特征,提出了一种三角网格平面法向量聚类的方法对零散点云进行特征提取,通过形状与颜色特征的双重匹配实现对环境中小型物体的识别。最后的实验结果证明,利用上述方法可以有效地利用三维激光点云数据和图像信息来实现对室内场景的三维重建,为未知环境下的移动机器人导航提供辅助。

A 3D laser scanner was built upon the pan-tilt and two-dimensional laser rangefinder. The depth data and image information from the indoor environment were fused at the pixel-level to create a 3D color point clouds model. Considering the poor visual expression of this model, a combination of RANSAC （Random Sample Consensus） and 1CP （lterative Closest Points） was used to extract the plane frame features from the laser point clouds. A triangular mesh normal vector clustering algorithm was proposed to extract the shape and size features from the scattering point clouds, combining with the color image information to recognize the small objects in the scene. The final experimental results show that the 3D laser point clouds and image information can be effectively processed to achieve the 3D reconstruction of indoor scenes. This work can provide the accurate environmental information for the navigation of mobile robot in unknown environment.