Computationally Efficient Hough Transform for 2D Object Location

This paper discusses the use of Hough transforms for feature-based object location. Though the GHT is highly robust and operates in polynomial (O(n2)) time, in real-time applications further reductions in computation are often required. A feature-orient ated HT has been designed which offers further speedup by factors of ~3, though robustness is also reduced. However, if objects are not immediately recognised, the new scheme can be made to revert gracefully to the normal GHT. Loss in robustness is most severe when objects exhibit a high degree of internal symmetry, but for objects with seven or more features arranged nonsymmetrically the new scheme should be especially valuable.