A new edge-grouping algorithm for multiple complex objects localization

We present a new algorithm that provides an efficient localization method of elliptic industrial objects. Our proposed feature extraction inherits edge grouping approaches. But instead of utilizing edge linkage to restore incomplete contours, we introduce criteria of feature's parameters and optimize the criteria using an extended Kalman filter. Through a new parameter estimation under a proper ellipse representation, our system successfully generates ellipse hypotheses by grouping the fragmental edges in the scene. An important advantage of using our Kalman filter approach is that a desired feature can be robustly extracted regardless of ill-condition of partial edges and outlier noises. The experiment results demonstrate a robust localization performance.