A KD curvature based corner detector

Due to the low efficiency of curvature calculation in corner detection algorithms, we propose a new corner detection technique for discrete curvature estimation based on KD curvature. Firstly, KD curvature is redefined and computed for each point on the curve after edge extraction and smoothing. Then, non-maximum suppression is used for obtaining candidate corner sets. Finally, the refined corner sets are retained with false and unstable corners removed. In addition, we introduce corner strength, as a new concept, for controlling detection precision. Our experimental results show that the proposed method outperforms existing detectors in both computational efficiency and flexibility of corner detection. Moreover, the average repeatability rate and local error rate are raised about 20 percent respectively.