Curve Detection from Images in Video-Based Navigation Methods

Abstract

This study investigates image-based navigation methods and algorithms for devices, with a focus on curve detection as a key step in object recognition. Image-based navigation relies on identifying objects in the environment and determining the device’s position based on their coordinates. Roads were selected as the objects of interest, and their identification was explored through curve detection in images. Curve detection is a fundamental problem in computer vision, as curves represent object boundaries, contours, and trajectories, and play a critical role in shape analysis, cartography, and navigation systems. The study employs the Hessian matrix, constructed from second-order derivatives of pre-processed images, to analyze local curvature. By examining the eigenvalues of the Hessian matrix, local structures are classified, and pixels belonging to curves are identified with high precision. A curvature detection criterion based on the relative magnitudes of the eigenvalues is applied: if one eigenvalue is significantly larger than the other, the corresponding pixel is determined to lie on a curve. This approach enables accurate and robust identification of road structures in images, forming a reliable foundation for subsequent navigation and mapping applications.