The reconstruction of realistic 3D models from digital images is one of the main issues of photogrammetry and computer vision. In practice, it is exquisitely hard to generate a realistic model with available technologies. Detection operators are the first step in a measure process, which allows finding points useful for image matching. For this primitives are used such as lines or points. Common corners detection methods are the Moravec and Harris operators. They can be applied to detect characteristic points of objects, such as corners. However, the operators are determined by many parameters such as the threshold or window parameters, that should be adjusted for each image separately. In this paper the method that allows better localization of corners, by using an edge detection, is presented. The method enables detection of corners on the edges, that limits the influence of the corner detector’s parameters. Two forms of edge detection were tested: sharp and soft. In the first case the image is transformed to a binary form. In the second, the edge points assume different intensities. The selection of an algorithm affects in a significant way the number and distribution of the detected corners. All the strategies, i.e. detection on the original image and with soft and sharp detection of the edges were compared for pictures with varied illumination and objects with different types of facture. Experiments that have been done allow development of an automatic feature detection and an image measure, which constitutes the main process that aims at creating procedures for an automatic detection of a cloud of points with high accuracy in multi-image sets used to generate a realistic 3D model of a measured object.