Abstract: According to the problem of determining the next best measurement posture in automatic measurement of robot vision,a method of fusion depth image and density clustering was proposed.The depth image and 3D point clouds of the object under the initial view were obtained by using the structured light binocular vision measurement technology.Then the edge and density clustering area of the object could be gained quickly by depth image.The complexity of each edge region of the object was determined based on density clustering,and the weight of each sub-region was calculated in combination with the size of the field of view,so the best moving direction of the next field of view was estimated.Furthermore,trend surface analysis was adopted to predict the spatial range of the next best measurement position.Using the depth image information,the global area of the trend surface analysis and the spatial data of the central trend line could be quickly obtained to determine the next best measurement posture.The Universal Robot 5 (UR5) robot and the vision measurement platform was established to carry out the measurement experiment on the Huba and rabbit models,and the effectiveness of the proposed method was proved.

Key words: robot vision, automatic measurement, depth image, density clustering, next best measurement posture

摘要： 针对机器人视觉自动测量中的下一最佳测量位姿确定问题,提出一种融合深度图像与密度聚类方法。该方法采用结构光双目视觉测量技术获得初始视角下被测物体的深度图像和三维点云数据,基于深度图像信息快速获取被测物体的边缘与密度聚类分析区域,基于密度聚类方法判定区域的复杂程度,进一步结合视场要求计算子区域权重,获得下一视场的最佳移动方向。利用趋势面分析预测下一最佳测量位姿的空间范围,采用深度图像信息获得趋势面分析全局区域,并快速获得中心趋势线的空间数据以确定下一最佳测量位姿。建立了Universal Robot 5(UR5)机器人和视觉测量平台,对胡巴和海盗头像模型进行测量实验,验证了所提方法的有效性。

关键词: 机器人视觉, 自动测量, 深度图像, 密度聚类, 下一最佳测量位姿