A real-time visual monitoring method for landslide creep using platform attitude self-calibration

High-frequency, high-precision, real-time displacement monitoring is essential for early warning of landslide creep, a key precursor to slope failure. However, platform attitude disturbances, such as wind-induced vibration and foundation settlement, can introduce spurious displacement into image sequences and reduce the accuracy of conventional vision-based monitoring. This study proposes a real-time visual monitoring technique for landslide creep based on platform attitude self-calibration. An optical tracker and an inertial measurement unit were integrated to acquire target images and attitude data simultaneously. A displacement decomposition model incorporating attitude fluctuations was developed to separate true slope movement from attitude-induced spurious displacement. Subpixel target localization used gradient-direction voting and weighted elliptical fitting, and real-time attitude estimation used an error-state Kalman filter. Indoor vibration tests at 40 m achieved an RMSE below 0.7 mm, with reductions of 0.955 and 0.846 compared with vision-only and vision-plus-reference-point correction approaches, respectively. During a 93-day field campaign at the Dayangshan landslide, Suzhou, the approach achieved a data capture rate of 0.986, an RMSE of 0.28–0.36 mm, and a correlation coefficient above 0.91 versus GNSS data. The approach provides a practical option for early warning of slope failure by accurately and continuously monitoring landslide creep under unstable platform conditions and unfavorable weather.