Novel nonlinear adaptive composite filters for illuminationinvariant pattern recognition are presented. Pattern recognition is carried out with space-variant nonlinear correlation. The information about objects to be recognized, false objects, and a background to be rejected is utilized in an iterative training procedure to design a nonlinear adaptive correlation filter with a given discrimination capability. The designed filter during recognition process adapts its parameters to local statistics of the input image. Computer simulation results obtained with the proposed filters in nonuniformly illuminated test scenes are discussed and compared with those of linear composite correlation filters with respect to discrimination capability, robustness to input additive and impulsive noise, and tolerance to small geometric image distortions.

La conservación de ecosistemas en zonas costeras debe contemplarse dentro de un marco sustentable con las actividades antropogénicas. Es relevante cuantificar el impacto que generan agentes externos, por lo que el objetivo de este trabajo es implementar un método para el monitoreo de la degradación en superficie de playa y vegetación litoral adyacente. Fueron recabadas capturas de imágenes aéreas de zonas costeras protegidas, obtenidas periódicamente mediante un vehículo drone. Se integró un dataset que contempla todas las fases estacionales y distingue 5 clases para su monitoreo: Mangle, vegetación rastrera, arena, mar y cerro-planicie. Para la segmentación semántica se implementaron y compararon distintas arquitecturas de redes neuronales convolucionales (CNN) empleando aprendizaje transferido. Los resultados han sido robustos en la clasificación, alcanzando una precisión global del 93.9% y entre 89.9-95.8% en las clases individuales. En la métrica Intersección sobre Unión, IoU, el rango fue entre 86.6-92.7%. En la detección de cambio son utilizadas series temporales para el monitoreo de clases. Este método ha sido aplicado al caso de estudio de la playa Ensenada Grande en el Parque Nacional Archipiélago Espíritu Santo.

Nowadays, the deformation measurement in metal sheets is important for industries such as the automotive and aerospace industries during its mechanical stamping processes. In this sense, Digital Image Correlation (DIC) has become the most relevant measurement technique in the field of experimental mechanics. This is mainly due to its versatility and low-cost compared with other techniques. However, traditionally, DIC global image registration implemented in software, such as MATLAB 2018, did not find the complete perspective transformation needed successfully and with high precision, because those algorithms use an image registration of the type “afine” or “similarity”, based on a 2D information. Therefore, in this paper, a DIC initialization method is presented to estimate the surface deformation of metal sheets used in the bodywork automotive industry. The method starts with the 3D points reconstruction from a stereoscopic digital camera system. Due to the problem complexity, it is first proposed that the user indicates four points, belonging to reference marks of a “Circle grid”. Following this, an automatic search is performed among the nearby marks, as far as one desires to reconstruct it. After this, the local DIC is used to verify that those are the correct marks. The results show reliability by reason of the high coincidence of marks in experimental cases. We also consider that the quality of mark stamping, lighting, and the initial conditions also contribute to trustworthy effects.

Classical correlation-based methods for pattern recognition are very sensitive to geometrical distortions of objects to be recognized. Besides, most captured images are corrupted by noise. In this work we use novel nonlinear composite filters for distortion-invariant pattern recognition. The filters are designed with an iterative algorithm to reject a background noise and to achieve a desired discrimination capability. The recognition performance of the proposed filters is compared with that of linear composite filters in terms of noise robustness and discrimination capability. Computer simulation results are provided and discussed.

No abstract

No abstract

Estimation of distance from objects in real-world scenes is an important topic in several applications such as navigation of autonomous robots, simultaneous localization and mapping (SLAM), and augmented reality (AR). Even though there is a technology for this purpose, in some cases, this technology has some disadvantages. For example, GPS systems are susceptible to interference, especially in places surrounded by buildings, under bridges or indoors; alternatively, RGBD sensors can be used, but they are expensive, and their operational range is limited. Monocular vision is a low-cost suitable alternative that can be used indoor and outdoor. However, monocular odometry is challenging because the object location can be known up a scale factor. Moreover, when objects are moving, it is necessary to estimate the location from consecutive images accumulating error. This paper introduces a new method to compute the distance from a single image of the desired object, with known dimensions, captured with a monocular calibrated vision system. This method is less restrictive than other proposals in the state-of-the-art literature. For the detection of interest points, a Region-based Convolutional Neural Network combined with a corner detector were used. The proposed method was tested on a standard dataset and images acquired by a low-cost and low-resolution webcam, under noncontrolled conditions. The system was tested and compared with a calibrated stereo vision system. Results showed the similar performance of both systems, but the monocular system accomplished the task in less time.

No abstract