Noncontact vision-based impact force reconstruction and spatial-temporal deflection tracking of a flexible barrier system under rockfall impact

Abstract

Flexible barrier systems are significant for natural disasters (e.g., rockfalls, mudslides) prevention in mountainous areas. However, monitoring the dynamic behavior of flexible barrier systems subjected to rockfall impact is difficult in practice due to the unpredictability of disasters and the problem of easily broken contact-type sensors. Therefore, this paper proposed a noncontact vision-based methodology, in which deep learning and computer vision techniques are developed to reconstruct impact forces induced by falling rocks and to identify the spatial–temporal deflection of a flexible barrier system from remotely captured image sequences by a high-speed camera. The contribution of this paper lies in two aspects: (1) a lightweight neural network is trained to detect falling rocks from captured images for motion information extraction and impact force reconstruction; (2) a two-dimensional distribution map of velocity amplitude is constructed for tracking the full-field spatial deflection pattern of the flexible barrier system under rockfall impact, and multipoint temporal deflection curves of the studied system are also extracted for maximum elongation calculation. The effectiveness of the proposed method is validated by full-scale experiments of a three-span flexible barrier system under rockfall impact with energies of 250 kJ and 750 kJ. The results show that the reconstructed impact forces, tracked full-field spatial deflection pattern, and extracted multipoint temporal deflection curve by the proposed method agree well with reference values, verifying the correctness and robustness of the proposed method. The obtained results can be directly used for the reliable design and condition assessment of flexible barrier systems under rockfall disasters.