Landslide Detection Using Digital Image Correlation and Vegetation Segmentation

Landslide is one of the most destructive natural hazard that causes great loss of economy as well as human lives. Landslide is usually monitored by geological surveys, satellite-based monitoring or human based sensor. In this work, Digital Image Correlation (DIC) is applied for landslide monitoring. DIC is a computer vision approach that can be applied on camera images with a random forest-based image segmentation. The study combines near real-time motion detection through Fast Fourier Transform (FFT) combined with vegetation masking to filter any noise induced by vegetation growth. This method can enhance the accuracy of landslide change detection over complex mountainous terrains. Results indicate that DIC along with vegetation masking was able to correctly track the displaced regions, which has significantly improved DIC reliability by filtering vegetation-induced motion artifacts.