基于三维点云曲率突变的航空发动机进气道异物检测方法
2024,32(12):67-72
摘要：为了解决目前人工巡检航空发动机进气道时存在的准确性差,效率较低等问题,提出一种基于三维点云曲率突变的进气道异物检测方法。首先,分析进气道点云信息因异物侵入而产生的空间形状变化。其次,提出一种基于体素网络的点云数据压缩方法,根据深度相机的水平视场角和垂直视场角,将点云数据划分至相应的体素网格中并进行均化压缩。再次,计算点云数据在进气道深度方向上的曲率变化,利用迭代阈值法计算识别曲率突变点云的动态阈值,采用区域生长法将离散的异物点云聚类分割为整体的异物目标。最后,搭建了航空发动机进气道的模拟测试环境,对3种不同尺寸异物的识别准确率达到95.4%,定位精度达到1.46cm,实验结果验证了本文所提方法的准确性与高效性。
关键词：异物检测;点云压缩;目标分割;迭代阈值法;区域生长法
A detection method of foreign objects based on curvature mutation of a three-dimensional point cloud for the intake duct of aircraft engine
Abstract：In order to solve the problems of low accuracy and efficiency in manual inspection of aircraft engine intake ducts, a foreign object detection method based on curvature mutation of three-dimensional point cloud is proposed for intake ducts. Firstly, the spatial shape variation of the inlet point cloud information is analyzed for the invasion of foreign objects. Secondly, a voxel network-based point cloud data compression method is proposed to divide the point cloud into voxel grids based on the horizontal and vertical field angles of depth camera. The point cloud in each voxel grid is averaged to compress data. Thirdly, the curvature variation of point cloud data is calculated in the depth direction of the inlet. The iterative threshold method is used to calculate the dynamic threshold to recognize the point cloud of curvature mutation from foreign objects. The region growth method is used to cluster the discrete point cloud of foreign object into the whole foreign object, thereby detecting and locating foreign objects in the inlet. Finally, an analog test environment is constructed for the intake duct of aircraft engine, in which three different sizes of foreign objects are detected and located. The recognition rate and the locating accuracy for foreign object reache 95.4% and 1.46cm, respectively. The experimental results verify the accuracy and efficiency of the method proposed in this paper.
Key words：Foreign object detection; Point cloud compression; Object segmentation; Iterative threshold method; Regional growth method
收稿日期：2023-12-12
基金项目：国家自然科学基金面上项目(61973154)、国防基础科研计划重点项目(JCKY2022209B001)、江苏高校“青蓝工程”优秀青年骨干教师项目(2022)资助
