Currently, most rescue robots are mainly teleoperated and integrate some level of autonomy to reduce the operator’s workload, allowing them to focus on the primary mission tasks. One of the main causes of mission failure are human errors and increasing the robot’s autonomy can increase the probability of success. For this reason, in this work, a stair detection and characterization pipeline is presented. The pipeline is tested on a differential drive robot using the ROS middleware, YOLOv4-tiny and a region growing based clustering algorithm. The pipeline’s staircase detector was implemented using the Neural Compute Engines (NCEs) of the OpenCV AI Kit with Depth (OAK-D) RGB-D camera, which allowed the implementation using the robot’s computer without a GPU and, thus, could be implemented in similar robots to increase autonomy. Furthermore, by using this pipeline we were able to implement a Fuzzy controller that allows the robot to align itself, autonomously, with the staircase. Our work can be used in different robots running the ROS middleware and can increase autonomy, allowing the operator to focus on the primary mission tasks. Furthermore, due to the design of the pipeline, it can be used with different types of RGB-D cameras, including those that generate noisy point clouds from low disparity depth images.

In the Yucatan Peninsula, there is a rich diversity of mangroves, notably including Rhizophora mangle, Avicennia germinans, and Laguncularia racemosa. These mangroves contribute to the recovery of degraded natural areas caused by human activities. Additionally, they serve as natural habitats for various animal and plant species. Studies have highlighted the significance of preserving and restoring these species through traditional methods. More recently, the integration of remote sensing and deep learning techniques has allowed for the automated detection and quantification of mangroves. In this study, we explore the application of deep neural network techniques to address computer vision challenges in the field of remote sensing. Specifically, we focus on the detection and quantification of mangroves in remote image sensing, employing transfer learning and fine-tuning with three distinct deep neural network architectures: SegNet-VGG16, U-Net, and Fully Convolutional Network (R-FCN), with the latter two based on the ResNet network. To evaluate the performance of each architecture, we applied key evaluation metrics, including Intersection over Union (IoU), Dice Coefficient, Precision, Sensitivity, and Accuracy. Our results indicate that SegNet-VGG16 exhibited the highest levels of Precision (98.03%) and Accuracy (97.03%), while U-Net outperformed in terms of IoU(96.97%), Dice Coefficient (92.20%), and Sensitivity (96.81%).

Due to high rates of self-inflicted accidents, driving a car is a demanding activity that requires the driver to prepare both physically and mentally. In this sense, driving assistance systems play an important role in preventing distractions and road accidents. In this work, we detect and classify driver distractions using convolutional neural networks trained first with the “State Farm Distracted Driver Detection” and then with a custom database. Furthermore, with the use of the TensorFlow Lite library, the trained model is deployed on a mobile device for real-time use. The best trained model reaches an accuracy of 91% in training and 95% in validation, thus resulting in a highly discriminating model for each of the classes (distractions) to be classified. In other words, 10 different classes have been identified: one of them is “safe driving” and the other 9 correspond to those most responsible for accidents. With the system's implementation on the mobile device, the test accuracy is reduced to 90% however it correctly detects distractions regardless of the class to which they belong. The initial results demonstrated the possibility of adapting an ADAS in a mobile device to detect distractions and generate real-time audible voice alerts.

Debido a los altos índices de mortandad derivados de la conducción distraída, la cual es ocasionada por el desvío de la atención de la conducción segura, de la carretera y del entorno del conductor, es que surgen los Sistemas de Asistencia a la Conducción (ADAS, por sus siglas en inglés) como una forma de prevenir estos siniestros y asistir al conductor durante su trayecto. En el presente trabajo, se detectan y clasifican la conducción segura y nueve actividades distractoras en los conductores de automóviles mediante el uso de redes neuronales convolucionales, aplicadas primero a la base de datos “State Farm Distracted Driver Detection” y también a una base de datos propia. Además, con el uso de la biblioteca “TensorFlow Lite”, el modelo se implementa en una aplicación móvil para su uso en tiempo real. El modelo resultante logra una exactitud del 91% en entrenamiento y del 95% en su validación. Con la implementación móvil, la exactitud del modelo se reduce en un 5%, sin embargo, se logran detectar y diferenciar correctamente las distracciones y se demostró la posibilidad de adaptar un ADAS en un dispositivo móvil.

We present a methodology for high-resolution orthomosaic reconstruction using aerial images. Our proposal consists a neural network with two main stages, one to obtain the correspondences necessary to perform a LR-orthomosaic and another one that uses these results to generate an HR- orthomosaic, and a feedback connection. The CNN are based onwell known models and are trained to perform image stitching and obtain a high-resolution orthomosaic. The results obtained in this work show that our methodology provides similar results to those obtained by an expertin orthophotography, but in high-resolution.