仿人机器人视觉导航中的实时性运动模糊探测器设计

针对仿人机器人视觉导航系统的鲁棒性受到运动模糊制约的问题，提出一种基于运动模糊特征的实时性异常探测方法. 首先定量地分析运动模糊对视觉导航系统的负面影响，然后研究仿人机器人上图像的运动模糊规律，在此基础上对图像的运动模糊特征进行无参考的度量，随后采用无监督的异常探测技术，在探测框架下对时间序列上发生的图像运动模糊特征进行聚类分析，实时地召回数据流中的模糊异常，以增强机器人视觉导航系统对运动模糊的鲁棒性. 仿真实验和仿人机器人实验表明:针对国际公开的标准数据集和仿人机器人NAO数据集，方法具有良好的实时性（一次探测时间0.1s）和有效性（召回率98.5%，精确率90.7%）. 方法的探测框架对地面移动机器人亦具有较好的普适性和集成性，可方便地与视觉导航系统协同工作.

A Real-time Method for Motion Blur Detection in Visual Navigation with a Humanoid Robot

To address the problem about robustness of humanoid robot visual navigation due to motion blur, a real-time method of motion blur detection based on motion blur feature is proposed. The negative impact of motion blur on visual navigation is analyzed, the motion blur law is studied and a no-reference method is then used to measure the motion blur feature of images captured by the robot. An unsupervised method is employed to cluster the blur features of images in the time sequence in an detection framework for recalling the anomaly from observations. The purpose is to improve the robustness of visual navigation to motion blur. Simulation and experiment on humanoid robot verify that the proposed method is real-time (0.1s per detecting) and effective (recall: 98.5%, precision: 90.7%) for an open standard dataset and the dataset acquired by NAO. The detection framework of the proposed method is universal and can be integrate with a robot visual navigation system.