Corner detection and its parameters: position, model, and orientation are useful for many computer vision applications, such as object recognition, matching, segmentation, 3D reconstruction, motion estimation [2, 3, 4, 34.] indexing, retrieval, robot navigation and in our case, edge tracking from design geometric. This need has led to the development of a large number of corner detectors [1, 5, 6, 7, 8, 9, 10, 11, 12, 13.]. Other corner detection methods are described in [14, 15]. These detectors compete with each other in terms of location precision, accuracy, speed and the information they provide. The classification and orientation of the model are the most interesting information needed in the edge tracking process. For some of these approaches, the CRF (Corner-Response-Function) can be invariant in terms of scale, rotation or even affine transformations. literature to place our contribution in context. Attempting to simultaneously perform corner detection and description of its properties turns out to be a complex job. In contrast, decoupling these two operations into two distinct steps simplifies and creates efficient problem solving.A. Corner detection: A wide variety of corner detectors are presented in the literature. One of the first operators of interest was developed by Moravec (Moravec, 1977). Detection methods can be grouped into three categories: grayscale-based methods [1, 5, 11, 12, 13, 18, 36], edge-based methods [24, 25, 26, 27, 28] and parametric models. methods based on [30, 31, 32, 33]. 1) gray level corner detectors: a large number of detectors operate directly on gray level images without requiring edge detection which is an inadequate model for Texture spots and blurry images. .... middle of paper ......function defined by 4 angles including one equal to .• V-: junction defined by 2 angles different from .• T-: junction defined by 3 angles including one equal to .• X-: junction defined by 4 angles, such that the sum of each two consecutive angles is equal to . • Y-: junction defined by 3 different angles these properties have mathematical meaning. For example, the K- model means that the current corner is a reflection point or the axis oriented by the angle is a reflection axis cf(Fig 6).III. EXPERIMENTAL RESULTS The results obtained generally show good behavior of the proposed method, implemented in Matlab. Two types of images were used for the tests: synthetic images (figure 8) and natural images (figure 11). Tables 1, 2 show the errors for some angles with our measurement method.