A new algorithm for real-time ellipse detection

We present a robust ellipse detector based on robust edge detection and Hough transform for real-time applications. The algorithm consists of three steps: we first perform edge detection, we than use a rule-based approach to compute edge convexity, and finally we decompose the parameter space of the Hough transform to achieve computational efficiency. We present our experiments, which show indeed that the proposed detector is able to detect elliptic features with an excellent accuracy under various image conditions and achieve a high speed at about 10 frames per second on a Pentium-III 500 MHz PC.