Landslide Early Warning Based on Retaining Wall Damage Monitoring by Real-Time Video

The frequency and severity of landslides is increasing, resulting in significant damage to people and infrastructure. Landslide early warning systems (LEWS) are becoming a key in damage mitigation of landslides. This study proposed a framework for building an early warning system of cut-slope areas based on monitoring damage of retaining wall using real-time video. Assessment of slope stability relies on observation of damage conditions such as cracks or displacements in the retaining wall. The identification of displacements and cracks in the retaining wall will be detected by a digital camera system integrated with deep learning and image processing techniques within calculation procedures. The performance of the proposed system is assessed using the lab experiments. The accuracies of displacement measurement ranged from 84.0 % to 99.4 %. While deep learning model achieved mean Average Precision values ranging from 0.86 to 0.90, and F1 score values, as the harmonic mean of precision and recall of the deep learning models, belong to the range of 0.83 and 0.85 in identifying cracks, and the dimensions of cracks were determined with the accuracies between 85.0 and 98.8 %. The correlation between retaining wall damage and slope stability is further investigated using numerical simulations. Subsequently, establishing threshold values ​​for both the displacement and the width of cracks in the retaining wall, which enables an early prediction of the occurrence of landslides.