A geometrical morphology-enhanced computer vision approach for structural health assessment

Abstract

Images contain abundant valuable information about the health state of the photographed infrastructures. However, local defects are mostly detected in vision-based structural health monitoring (SHM), while global safety and risk at a larger scale are rarely assessed. To fill up this gap, a geometrical morphology-based image analysis framework is developed for structural global health assessment. A structured random forest edge detector is adopted to extract the edges in an image, and the morphological operations are subsequently used to highlight the edge skeleton, from which the continuous line segments are estimated by the Hough transform. Via these operations, the intrinsic geometrical features of different structures can be extracted and highlighted in an unsupervised manner and at a real-time speed. A global health indicator, line-to-edge ratio, is finally calculated to assess the structural state according to the fact that the deterioration of material will introduce abundant irregular edges in an image and break the lines at the structure boundaries. A destructive beam loading test video and a set of images from different bridge piers are analyzed for verification. The images are taken from arbitrary, uncontrolled, and backlighting views. The results show that the proposed framework can correctly quantify the health condition and detect the occurrence of damage for the in-field infrastructures.