Landslide detection with feature vectors extracted from video of fixed monitor

In Taiwan, landslides occur very often in typhoon seasons recently, especially after heavy rain. They usually cause serious damages and even some mortality. Therefore, it is very important to detect landslides for an early warning system to reduce the damages. In this study, we apply image processing techniques on video taken by still-camera to monitor landslides. Texture features, color histogram and color moment of landslides training events are extracted to form feature vectors and choose the best feature subsets by particle swarm optimization (PSO). PSO is an optimization method to find the best value by simulating birds search for foods. Then Fully Constrained Least Squares (FCLS) is applied to classify the feature vectors into the composition percentages of landslides. Finally, the decision of landslide is made by fuzzy decision method. Our propose method can be implemented to a computer aided system which can reduce the error causing by human.