Abstract: A natural landmark extraction method based on 2D laser rangefinder is described.The framework consists of three main parts: data clustering,filtering and feature extraction.According to the scale space theory in computer graphics,a curve-based estimator is developed using UKF(unscented Kalman filter),and the scan point topology describing the local environment is estimated.The filtering convolution kernel is constructed with the Mahalanobis distances obtained during estimation,and adaptive filtering of the original range image is achieved.Clustered data is segmented and characterized by the curvature function of the range data.This method is robust to noise,and can reliably extract natural landmarks in unstructured environments.Experimental results show that the proposed method is efficient in natural-landmark extraction,which can provide plenty landmarks for navigation system of autonomous mobile robot.