改进法线方向的点云实时分割提取平面方法研究

Real-time Point Clouds Plane Segmentation Based on Improved Normal Direction

为了提高服务机器人在室内环境中目标物体抓取的效率,提出一种针对三维点云数据改进表面法线估计的方法;重点研究了如何提高使用三维点云数据实时性和准确性,这是机器人抓取过程的关键因素;使用消费级别的传感器kinectV2获取点云数据,提出使用积分图像降低滤波区域内像素的遍历消耗的时间,从而提高实时性;采用动态调整平滑区域的大小以及分配权重的方法平滑处理点云数据的噪声,提高法线估计精度;依据改进后的法线方向,利用区域生长算法对不同指向的法线进行分类,然后对点云数据实施分割提取平面;最后,采用提出的方法和点云库中的相关算法,进行了平面分割提取测试比较;结果表明,利用kinectV2获取的深度数据(1-5米)比目前点云库中的平面分割算法数据精度更高、实时性更强。

In order to improve the efficiency of grasping objects by service robot in the indoor environment,we present a method for improving the surface normal estimation of point cloud.Real-time and accuracy are the key component of grasping objects by service robot.We get the point cloud by the consumer-level sensor kinectV2.It is proposed to use the integrated image to reduce the time of traversal consumption of pixels in the filter area for improving real-time.Smoothing the point cloud data dynamically adjust the size and assign weights for the smoothing area for improving the accuracy of normal estimation.According to the improved normal direction,the region growing algorithm is used to classify the normal in lines by different directions,and then the planes are extracted from the point cloud.Finally,a comparison of the plane segmentation tests was performed by our methods and the Point Cloud Library.The results show,our method is more accurate and real-time than the Point Cloud Library methods for the plane segmentation with the depth（1-5）meter of point cloud.