面向建筑健康监测的无人机自主巡检与裂缝识别

UAV Autonomous Inspection and Crack Detection Towards Building Health Monitoring

摘要:

面向建筑健康的高时效监测需求，为提升表面病害视觉检测的自动化水平，提出了场景信息引导的无人机自主巡检任务规划方法。利用场景先验信息，针对建筑结构特征，设计了平行观测和包络观测两种观测模式，实现了狭窄空间中独栋建筑的全覆盖避障巡检，获取了毫米级分辨率的整体建筑观测影像，并对巡检质量的整体评估提出了有效的量化指标。将建筑立面划分为3 720个子区域，利用深度残差网络开展了裂缝识别分类。错分13个子区域，漏分14个子区域，裂缝识别精度高。将裂缝骨架线映射到建筑三维模型，为裂缝形态与建筑整体的一体化表达提供了数据支撑。该研究将高精度三维重建与表面病害识别相结合，为建筑健康一体化监测提供了有效的观测分析手段。

Abstract:

Aiming at the demands of time-sensitive building health monitoring to promote the automation level of surface disease visual inspection， scene information guided UAV inspection mission planning was proposed. Based on the scene’s prior information， two observation modes， parallel observation and envelope observation， were designed for the structural characteristics of the building， which realized the full coverage obstacle avoidance inspection of the individual building in the narrow space as well as the observation of whole building with millimeter resolution. Meanwhile， a series of effective quantitative indexes for the overall evaluation of the inspection quality were put forward. The facade of the building was divided into 3 720 subregions. The surface cracks were identified and classified by a deep residual network. The result shows that 13 wrong subregions and 14 missing subregions reflect the high accuracy of crack identification. The crack skeletons are mapped to the reconstructed 3D model， which provides data support for the integrated expression of crack morphology and building information. This study combines high-precision 3D reconstruction with surface disease recognition， providing a practical observation and analysis method for integrated building health monitoring.