Real-time obstacle avoidance using an MPEG-processor-based opticflow sensor

摘要

Describes a vision system for obstacle detection in mobile robotnavigation. The system uses an image processing board equipped with anMPEG motion estimation processor that calculates a robustoptic-flow-like vector field in real-time. This field is then evaluatedby algorithms running in software on the host PC. As the solutions tothe general problem of structure and motion from optic flow are tooinstable for the use in this application, the typical constraints ofmobile robotics are exploited, i.e. a reduced set of motion parametersand a known ground plane. Ego-motion can then be reconstructed withrobust one dimensional methods. A new criterion for obstacles that copeswell with the noise properties of the motion field is introduced. Forvectors belonging to obstacles the 3D information is reconstructedallowing not only qualitative detection of obstacles but quantitativepath planning