WE INTRODUCE AN ALGORITHM TO CHARACTERIZE SURFACES IN RANGE DATA THROUGH THE USE OF DIFFERENTIAL GEOMETRY DESCRIPTORS. OUR METHOD IS ORIGINAL IN THE SENSE THAT THE CHARACTERIZATION IS DONE IN A ONE-PASS, CONVOLUTION-LIKE PROCESS FOR WHICH NO ORTHOGRAPHIC PROJECTION IS REQUIRED. THIS APPROACH ALLOWS AN ARBITRARY SAMPLING MOVEMENT OF THE RANGE IMAGING SENSOR, A PROPERTY THAT CAN BE ESSENTIAL IN REAL-TIME APPLICATIONS. WE COUPLE THE CHARACTERIZATION ALGORITHM WITH A SEGMENTATION STRATEGY BASED ON A GENERALIZATION OF BLOB COLORING FOR BINARY IMAGES, USING PROXIMITY AS THE CONNECTION CRITERIA. THE VALIDITY OF OUR APPROACH IS TESTED IN AN EXPERIMENT DESIGNED TO LOCALIZE AND COLLECT CANS IN A STRUCTURED ENVIRONMENT, USING A MOBILE ROBOT EQUIPPED WITH A LASER RANGE FINDER.WE INTRODUCE AN ALGORITHM TO CHARACTERIZE SURFACES IN RANGE DATA THROUGH THE USE OF DIFFERENTIAL GEOMETRY DESCRIPTORS. OUR METHOD IS ORIGINAL IN THE SENSE THAT THE CHARACTERIZATION IS DONE IN A ONE-PASS, CONVOLUTION-LIKE PROCESS FOR WHICH NO ORTHOGRAPHIC PROJECTION IS REQUIRED. THIS APPROACH ALLOWS AN ARBITRARY SAMPLING MOVEMENT OF THE RANGE IMAGING SENSOR, A PROPERTY THAT CAN BE ESSENTIAL IN REAL-TIME APPLICATIONS. WE COUPLE THE CHARACTERIZATION ALGORITHM WITH A SEGMENTATION STRATEGY BASED ON A GENERALIZATION OF BLOB COLORING FOR BINARY IMAGES, USING PROXIMITY AS THE CONNECTION CRITERIA. THE VALIDITY OF OUR APPROACH IS TESTED IN AN EXPERIMENT DESIGNED TO LOCALIZE AND COLLECT CANS IN A STRUCTURED ENVIRONMENT, USING A MOBILE ROBOT EQUIPPED WITH A LASER RANGE FINDER.