基于PointNet++的煤场点云分割与识别方法

Point cloud segmentation and recognition method of coal yards based on PointNet++

为了实现煤场环境下的实时监控与安全监测,对煤场环境应用了一种基于PointNet++的目标分割与识别的方法。利用二维激光扫描仪做直线运动的装置采集三维点云数据,通过设置目标安全距离,采用基于欧氏距离的点云分割算法对原始点云进行分割,调用训练好的PointNet++网络对分割后的目标点云进行识别,对识别结果进行判断,并分析目标物体的工作状态是否安全。实验结果表明:煤场环境典型物体点云的分割精确率与召回率均大于90%,目标识别准确率达到98%,验证了基于PointNet++点云分割与识别方法的可行性。

In order to achieve real-time monitoring and safety monitoring in the coal yard environment,a target segmentation and classification identification method based on PointNet++is applied to the coal yard environment.First of all,the use of two-dimensional lidar to do linear motion to collect three-dimensional point cloud data,by setting the target safety distance,the use of Euclidean distance-based point cloud segmentation algorithm to split the original point cloud,and then,through the trained PointNet++network to identify the split target point cloud,and finally the target monitoring results to determine whether the target working state is safe.Experimental results show that the segmentation accuracy and recall rate of typical objects in the coal yard environment are greater than 90%,and the final classification recognition accuracy reaches 98%,which verifies the feasibility of the PointNet++segmentation and recognition method.