基于Scheimpflug成像的大景深三维视觉系统及目标识别研究

In this project, we study 3D vision imaging technology with large depth of field based on Scheimpflug principle and 3D object recognition technology. A tilt lens imaging system based on Scheimpflug principle is proposed to overcome the problem of limited depth of field in 3D measurement, and new scientific issues are studied caused by tilt lens imaging based on Gaussian optics: versatile camera model for cameras with tilt lenses under Scheimpflug principle, high precision calibration for Scheimpflug optical 3D vision imaging system and its reconstruction theory. Based on the 3D point cloud, 3D object recognition and localization technology with global modeling and local feature matching is studied. A global description model based on the directional point pair feature is established, and a fast voting matching mechanism based on the generalized Hough transform principle is designed to realize the rapid recognition and localization of object using sparse point clouds. The research results of this project will enrich and supplement the visual measurement theory and promote the development of large depth of field 3D imaging and object recognition technology, which plays an important role in the field of robot 3D visual guidance and 3D object tracking.

本项目研究基于Scheimpflug原理的大景深3D视觉成像及3D目标识别技术。提出构建基于Scheimpflug原理的倾斜镜头成像系统解决三维测量中景深受限的问题，并基于高斯光学研究因镜头倾斜成像而带来新的科学问题：Scheimpflug条件下倾斜镜头成像的通用摄像机模型、Scheimpflug 3D视觉成像测量系统的高精度标定以及立体重构技术；在获得物体3D点云数据的基础上，研究基于全局建模和局部特征匹配的3D目标识别和定位技术，建立基于方向点对特征的物体全局描述模型，设计基于广义霍夫变换原理的快速投票匹配机制，实现稀疏点云目标的快速识别和定位。本项目的研究成果将丰富和补充视觉测量理论，推动大景深3D成像和3D目标识别技术的发展，并在机器人3D视觉引导、3D目标跟踪等领域具有重要的应用价值。

本项目研究基于Scheimpflug原理的大景深3D视觉成像及3D目标识别与定位技术。针对三维测量中景深受限的问题，提出并构建了基于Scheimpflug原理的倾斜镜头成像系统，在此基础上利用厚透镜投影特性分析建立了Scheimpflug 条件下任意倾斜镜头的摄像机通用成像模型。相比目前提出的薄透镜成像模型，本模型可描述各种光路系统的倾斜成像和光轴相对成像面任意倾斜的情况，具有广泛的适用性。并基于该模型，研究提出了Scheimpflug 3D视觉成像测量系统的高精度标定以及立体重构方法；在获得场景的3D点云数据的基础上，开展了基于全局建模和局部特征匹配的3D目标识别和姿态估计技术，提出了高鲁棒性高重复性的3D点云局部参考坐标系建立方法以及基于空间和几何信息的点云局部特征描述符，实现了稀疏点云目标的快速识别和定位。本项目的研究成果丰富和补充视觉测量理论，推动大景深3D成像和3D目标识别技术的发展，并在机器人3D视觉引导、3D目标识别与姿态估计等领域具有重要的应用价值。