基于点云数据实时配准算法的桥梁结构复杂缺陷形态识别

Complex defect morphology identification of bridge structures based on point cloud data real time registration algorithm

桥梁结构损伤检测过程中,大多依托于二维图像完成结构复杂缺陷的形态识别,只考虑表面的灰度和纹理等信息,使得识别结果的平均精度均值(mAP)较低。因此,提出基于点云数据实时配准算法的桥梁结构复杂缺陷形态识别方法。首先采用激光扫描仪和传感器设备,采集桥梁结构三维点云数据,并运用基于密度的聚类算法进行点云数据聚类分割,实现点云数据去噪处理;然后利用柔性动作-评价(SAC)配准算法、改进迭代最近点(ICP)配准算法进行点云数据实时配准,考虑深度、高度等第三维属性完成复杂缺陷区域检测;最后针对缺陷区域的测量点分别计算结构位移函数,基于此识别出缺陷的具体形态。试验结果表明,所提方法得出的复杂缺陷形态识别结果的mAP值大于0.92,基本满足了桥梁检测要求。

In the process of bridge structural damage detection,most of the recognition of complex defect morphology relies on two-dimensional images,only considering information such as surface grayscale and texture,resulting in low Mean Average Precision(mAP)of the recognition results.Therefore,a real-time registration algorithm based on point cloud data was proposed for the recognition of complex defect morphology in bridge structures.Firstly,laser scanners and sensor equipment were applied to collect three-dimensional point cloud data of bridge structures,and density-based clustering algorithms was used for point cloud data clustering and segmentation to achieve point cloud data denoising processing.The Soft Assignment Cost(SAC)registration algorithm and improved Iterative Closest Point(ICP)registration algorithm were used for real-time registration of point cloud data,complex defect area detection was completed by considering three-dimensional attributes such as depth and height.Finally,the structural displacement function separately for the measurement points in the defect area was calculated,and the specific defect morphology was identified based on this.The experimental results showed that the mAP value obtained from the application of the proposed method for complex defect morphology recognition was higher than 0.92,which basically met the requirements of bridge detection.