面向工业机器人的三维视觉感知与引导关键技术

3D vision perception and guidance is an effective approach for personalized and flexible manipulation. Based on in depth analysis on the state-of-the-art the shortcoming of the key technology of nowadays 3D vision perception and guidance, and based on also on our previous works, in this project we propose some creatives to tackle some key problems in the 3D vision perception and guidance of industrial robot. The project consists of 3D vision calibration and 3D working environment modelling, 3D object detection and recognition, tracking and localization, as well as the 3D vision based guidance and control of industrial robot. The creatives include: aiming at high precise 3D environment modelling of different surface, we propose the binary structured light strategy; aiming at the problem of object obstacle and stack we propose the subspace and graph based 3D object detection and recognition method; aiming at real-time 3D localization and tracking we propose the model based method; aiming at handling both flexible manipulation and real-time requirement, we propose the offline+online learning method for industrial robot guidance and control. In addition to theoretical and algorithmic achievements, this project will also construct the industrial robot prototype platform. The success of the project will improve the technology of 3D vision perception and guidance of industrial robot in theory, model, algorithm and also application.

三维视觉感知和引导是实现工业机器人个性化柔性生产的有效手段。本项目在对工业机器人三维视觉感知和引导关键技术的现状和瓶颈深入分析基础上，结合自身前期基础，突破关键技术，提出创新解决方案。项目主要研究内容包括工业机器人三维视觉标定和三维工作环境建模、三维目标检测与识别、定位与跟踪，基于三维视觉的工业机器人引导与控制等技术。创新性工作包括：针对不同反射强度表面的高精度鲁棒三维环境建模，提出创新的二值结构光编码策略等；针对目标的遮挡和堆积问题提出了基于子空间模型和基于结构图模型的三维目标检测识别方法；提出基于模型的实时三维位姿测量和跟踪方法；兼顾个性化柔性生产和实时化需求提出新型在线学习+离线学习的机器人三维视觉引导和控制框架。除创新性理论算法成果，本项目还将研发工业机器人操作原型系统平台。项目的成功实施将从理论、模型、算法和应用几个层面提高工业机器人三维视觉感知和引导的技术水平。

三维视觉感知和引导是实现工业机器人个性化柔性生产的有效手段。本项目在对工业机器人三维视觉感知和引导关键技术的现状和瓶颈深入分析基础上，结合自身前期基础，突破关键技术，提出了创新解决方案。项目主要研究了三方面内容，1）三维视觉标定和三维工作环境建模等共性底层技术；2）工业机器人三维目标检测与识别、定位与跟踪，3）基于三维视觉的工业机器人引导与控制。创新性成果包括：1）针对不同反射强度表面的高精度鲁棒三维环境建模，提出了创新的二值结构光编码策略等；2）针对目标的遮挡和堆积问题提出了基于子空间模型和基于结构图模型的三维目标检测识别方法，基于深度学习的实时三维位姿测量和跟踪方法；3）兼顾个性化柔性生产和实时化需求提出了在线学习+离线学习的机器人三维视觉引导和控制框架。除创新性理论算法成果，本项目还研发了基于三维视觉感知的工业机器人操作原型系统平台。