Marcelo Petry

Abstract
Optical character recognition (OCR) is the process by which the textual content of an image is converted into strings. Localization is the problem of figuring out where one is in a given environment. In this work we approach the application of OCR in robot localization. We develop and test a vision based localization system that is capable of detecting room identification signs present in the environment, recognizing their textual contents and apply them to determine its location referent to a topological map of the environment. A sign detection method based on image segmentation by color and corner detection by contour analysis is developed. The recognition of characters is performed with the application of an open-source OCR engine. Localization is performed through the comparison of sign readings with the textual information embedded in the topological representation of the environment. The algorithm was tested in a dataset of images acquired in a corridor. Experimental results show that the system successfully determines its localization in 83.33% of tested cases. © 2019 IEEE.

Abstract
The photogrammetry, 3D reconstruction from images, is an old technique but it's potentials could only be seen after the development of computers and digital photographs. Nowadays it has many applications, as creating scenarios for games, acquiring human expressions, roof inspection, stockpile measurement, high voltage transformer inspection, etc. As new technologies appear, new applications to photogrammetry are created. In this paper the use of available open and closed-source algorithms for 3D reconstruction and texturization is investigated. To achieve this goal, images of a fountain from several points-of-view were used. Next a comparison between several open and closed-source algorithms was performed, evaluating the number of faces, time consumption, RAM memory, GPU memory and the generated textured 3D models. The results obtained demonstrate that with the right setup, current open-source algorithms can achieve results near or better than proprietary software. Regarding the comparison, 3Dflow and MeshRecon presented the most accurate textured 3D models. When comparing quantitative measures, though, MeshRecon presented a slightly better performance in time consumption, but 3Dflow had a better RAM memory usage and a lower quantity of faces with a similar level of details. © 2019 IEEE.

Abstract
Inspection of railroad tracks is still predominantly performed visually by human inspectors. Due to the extension of the tracks this is a slow and tedious operation, significantly subjected to human errors and inconsistency. In this context, computer vision systems, composed of field-acquired images and processing algorithms, have a great potential to improve efficiency and to offer systematic inspection methodologies. In this paper the use of available point cloud and mesh generation algorithms to construct 3D surface of railroad tracks is investigated. To achieve this goal, images of a small track were acquired from several points-of-view. Next a comparison between several open and closed-source algorithms was performed, evaluating the number of 3D points, time consumption, RAM memory, GPU memory, number of faces, and the generated mesh. The results obtained demonstrate that with the right setup, current image processing methodologies can be used to construct 3D surfaces of uncontrolled scenarios, such as those of a real railroad environment. Regarding the comparison, SURE and Poisson presented the most accurate meshes. When comparing quantitative measures, though, Poisson presented a slightly better performance in time consumption, but SURE had a better RAM memory usage and a greater level of details. © 2017 IEEE.

Abstract
Nowadays, there are many technologies that support location systems involving intrusive and nonintrusive equipment and also varying in terms of precision, range, and cost. However, the developers some time neglect the noise introduced by these systems, which prevents these systems from reaching their full potential. Focused on this problem, in this research work a comparison study between three different filters was performed in order to reduce the noise introduced by a location system based on RFID UWB technology with an associated error of approximately 18 cm. To achieve this goal, a set of experiments was devised and executed using a miniature train moving at constant velocity in a scenario with two distinct shapes-linear and oval. Also, this train was equipped with a varying number of active tags. The obtained results proved that the Kalman Filter achieved better results when compared to the other two filters. Also, this filter increases the performance of the location system by 15% and 12% for the linear and oval paths respectively, when using one tag. For a multiple tags and oval shape similar results were obtained (11-13% of improvement).

Abstract
Most of the original image feature detectors are not able to cope with large photometric variations, and their extensions that should improve detection eventually increase the computational cost and introduce more noise to the system. Here we extend the original SURF algorithm increasing its invariance to illumination changes. Our approach uses the local space average color descriptor as working space to detect invariant features. A theoretical analysis demonstrates the impact of distinct photometric variations on the response of blob-like features detected with the SURF algorithm. Experimental results demonstrate the effectiveness of the approach in several illumination conditions including the presence of two or more distinct light sources, variations in color, in offset and scale.