Real-Time Feature Extractio: a Fast Line Finder for Vision-Guided

摘要

THE SHEER AMOUNT OF DATA CONTAINED WITHIN AN IMAGE CAN POTENTIALLY MAKE PIXEL-BASED ALGORITHMS COMPUTATIONALLY INFEASIBLE. THERE ARE TWO WAYS TO IMPROVE THE SPEED OF THESE ALGORITHMS: THE AMOUNT OF EFFORT EXPENDED ON EACH PIXEL CAN BE REDUCED BY CAREFUL CODE OPTIMIZATION, AND, THE NUMBER OF PIXELS FULLY PROCESSED BY THE ALGORITHM CAN BE REDUCED. REDUCING THE EFFORT EXPENDED ON EACH PIXEL REDUCES THE TIME REQUIRED TO PROCESS AN IMAGE BY A CONSTANT FACTOR. SELECTIVE PROCESSING, SUCH AS FOCUS OF ATTENTION, CAN DE- CREASE OVERALL COMPUTATION BY SEVERAL ORDERS OF MAGNITUDE BY EXCLUDING IR- RELAVANT PIXELS WHICH DO NOT SIGNIFICANTLY CONTRIBUTE TO THE FINAL RESULT. THIS PAPER DEVELOPS A FAST PIXEL-BASED ALGORITHM WHICH USES THESE PRIN- CIPLES TO ACHIEVE REAL-TIME FEATURE EXTRACTION OF LINES FOR USE IN VISION- GUIDED MOBILE ROBOT NAVIGATION. IT IS BASED UPON A LINE EXTRACTION ALGORI- THM FIRST DEVELOPED BY BURNS ET.AL. [3], THOUGH IT DIFFERS SIGNIFICANTLY IN THE WAY PIXELS ARE PROCESSED, LINES ARE FITTED, AND ITS INHERENT TIME PERFORMANCE. AT THE EXPENSE OF ROBUSTNESS AND RELIABILITY, THE ALGORITHM IS MODIFIED AND SIMPLIFIED SO THAT IT IS SIGNIFICANTLY FASTER. THE RESULTING FAST LINE FINDER (FLF) PROGRAM ALLOWS PARAMETRIC CONTROL OF COMPUTATIONAL RESOURCES REQUIRED TO EXTRACT LINES WITH PARTICULAR CHARACTERISTICS. AS A ROUGH COMPARISON, THE FLF PROGRAM RAN IN ABOUT TWO SECONDS FOR A 256 X 256 IMAGE ON A DEC VAX-11/750 VERSUS SEVERAL MINUTES REQUIRED FOR THE ORIGINAL