In view of the structural deformation problems of large bridges during the construction phase and operation period, currently, there isn’t automated, high-frequency, real-time , long-term and accurate detection means. Based on the theory of digital image processing and deep learning, a non-contact detection method suitable for large-scale bridge structure deformation is proposed. With this method, the system can synchronously and dynamically monitor multiple target structures of bridges. The method first obtains the dynamic video sequence image of the bridge structure through high-resolution photography equipment; secondly, preprocesses the high-definition image; then extracts the image ROI to determine the specific bridge structure to be processed; finally, combined with YOLO V3 algorithm in deep learning and improved surf algorithm to realize deformation detection of bridge structures. The experimental results show that the detection speed of the algorithm is between 20fps and 30fps, and the detection accuracy of the algorithm is within 0.2mm when the target distance is 100m, high detection accuracy, which effectively reflects the deformation of the bridge structure.

CICTP