Image Processing for Visual Navigation of Roadways

摘要

In support of the Autonomous Land Vehicle (ALV) project, which is a part of the Strategic Computing Program development by DARPA, the Computer Vision Laboratory of the University of Maryland has designed a visual navigation system which will enable the ALV to perform road following tasks. Road following is an example of intermediate range navigation. The task is to compute a path in an intermediate range environment through a region of uniform visibility/navigability established first by long range navigation. Road following consists primarily of three Vision Modules which embody image processing tools, geometrical and model-based constraints, and rule-based reasoning capabilities; it is the responsibility of these modules to establish a representation of the three-dimensional structure of the road scene. This system also includes a Planner, a Navigator and a Pilot. The Planner establishes goals and operating constraints for the Navigator, which is responsible for generalized path planning. The Pilot interprets this path into basic commands understandable by the ALV's motor controllers. The Vision Executive is the module responsible for the overall vision process control. It is also the Vision Modules and Executive which establish a labeling of the scene, consistent with: the representations derived from the image domain (2-D symbolics), the geometricial interpretations, the knowlege base.