An adaptive smoothing algorithm within a scale space framework is proposed to extract the features of noisy range image of a laser rangefinder. The method is composed of feature estimation and multi-scale smoothing. A Unscented Kalman Filter(UKF) is used to construct an adaptive feature estimator to estimate the topology of points,then the Mahalanobis distances obtained by estimation are taken to calculate the smoothing mask. In order to provide a more efficient estimation of different major geometries by a single model, an adaptive curve model which varies depending on the local nature of range image is employed. Experimental results indicate that the Peak Signal-to-Noise Ratio (PSNR) gain of the adaptive algorithm has reached 10.55 dB and the Mean Square Error(MSE) has been reduced by 58.24% when the noise variance is 2.25×10-4 m2. The proposed method with a adaptive model can improve the correctness of feature extraction by 10% comparing to the smoothing algorithm with a fixed neighborhood model, while the time consuming is reduced by about 55%.