A vision system which uses 3D sensory information as well as image intensity as input data for object recognition algorithms is a major component of a mobile robot guidance system. The vision system uses a special 3D range-finder that generates registered intensity and range images. From these images the fold, jump, and reflectivity edges in the scene are detected. The different edge types are fused into one edge image, forming the base for an edge-based segmentation technique. The resulting segments in the scene are then grouped into object regions, each likely to come from the same object. Hypotheses for objects are formed for each object region based on constraints on surface features and constraints between surfaces. Each hypothesized object's location and orientation is estimated and the hypotheses are verified. The verification is aided by an accurate simulator of the real range-finder. The simulator, using ray-tracing techniques, provides the range and intensity images that would be generated by the real range-finder. This verification can be done using only the sensed data. Furthermore, the simulator can be incorporated in an object tracking system necessary for a mobile robot. The initial hypotheses are updated as the robot moves through the environment, and the simulator can be used to generate the expected scene from each new mobile robot position. This simulated data can also be used to segment the real range-finder data in a model-driven manner and to select areas of the scene for selective scanning and processing.
展开▼
