Three-Dimensional Feature Extraction.

摘要

Abstract : Range images offer a significant advantages over passive reflectance images because they preserve the 3-D information of the scene viewed from the sensor. Therefore, range data is becoming an increasingly important source of information for a variety of applications including 3-D target classification, autonomous vehicles, and robot vision. This research is part of an effort to develop a 3-d object recognition system for vehicle objects in air-to-ground laser range imagery. The full system includes image feature extraction, object modeling, model-driven prediction, and feature to model matching. This paper presents several three-dimensional feature extraction techniques for use on laser range imagery. These include object-ground segmentation, projection image generation from range data, and 3-D physical edge detection. We emphasize extracting 3-D physical features of the object from 3-D range data without restricting ourselves in a sensor-centered range image format. The object-ground segmentation and projection image generation techniques extract global object features from range data, and are useful for object orientation estimation and major structures identification. The 3-d physical edge detector directly calculates the physical angle of the object surface. It is not only useful for physical edge (convex, concave, occluding) detection, but also provides useful information for extracting planar and curved surfaces.