Abstract
The core of the contour-based corner detection is essentially performing a good curvature estimation on planar curves. Inspired by intuitive observation that the curvature of a point on a contour is proportional to the distance accumulation of its neighbors to the tangent of the point, we present a novel curvature estimator named Relative Tangent-to-Point Distance Accumulation (RTPDA) for contour-based corner detection. In the approach, we fit the curve segments with quadratic polynomials by employing least square technique to derive the tangent of the target point, and then accumulate the distance of its neighbors to the tangent, which is a good approximation of the discrete curvature. Experiments verify the effectiveness and the efficiency of the proposed detector in comparison with several influential corner detectors under three commonly used evaluation metrics, namely, Average Repeatability (AR), Localization Error (LE) and Accuracy index (ACU).
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have