IMAGE SEGMENTATION USING LASER-RADAR DATA

In this work, we study ladar images of man-made objects in outdoor scenes. Our objective is to separate man-made objects from background. We explore ways to segment images from different modalities of ladar and integrate results to improve segmentation. We use planar surface fitting to segment the range image. The background usually cannot be fit into planar segments while man-made objects usually yield planar segments. The intensity image is segmented by using image statistics. Both surface fitting and statistical methods can be applied to the velocity component for segmentation. We combine these segmentation maps to generate a composite segmentation map. The final result shows strong resemblance to manual segmentation. These results can be fed into a cuing system for further processing, or into a higher level, knowledge-based system for scene interpretation. © 1990.