Computer vision for real-world imagery normally consists of threestages: acquisition/low-level feature extraction (e.g. capture followedby edge detection), medium-level feature extraction (typically into ageometric and/or topological space) and task-oriented sceneunderstanding (e.g. aggregation of geometric features to characteriseobjects of interest). The Hough transform (HT) is an efficient mediumlevel method to extract geometric features from an image which worksfairly well for images that contain noise and occlusion. However, itsperformance decreases with image and parameter space quantisation noise.This paper describes two HT variants based on an analytical leastsquares refinement procedure that helps overcome some of thesedifficulties. A parallel implementation on a transputer based system isalso discussed and evaluated
展开▼