GTGabriel Alejandro Tarnowski

Gabriel Alejandro Tarnowski

*This is an algorithmically generated author record based on publicly available information.

Gabriel Alejandro Tarnowski has a total of 10 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Improving ORB-SLAM3 Accuracy in Dynamic Scenes with YOLO11 Segmentation

    by Renata Raffaine Villegas, Anselmo Rafael Cukla, Gabriel Alejandro Tarnowski, Guillermo Mudry, Sergio Omar Lapczuk, Ely Carneiro de Paiva, Daniel Fernando Tello Gamarra

    Traditional Visual SLAM systems, like ORB-SLAM3, often lose accuracy in dynamic environments. This work presents YOLO11-ORB-SLAM3, an enhancement to ORB-SLAM3 for dynamic scenarios, which integrates a YOLO11-based instance segmentation module to detect and exclude dynamic features from the tracking process. The system is designed to work with stereo and RGB-D cameras, and its performance was evaluated on challenging dynamic sequences of the public TUM RGB-D dataset, and also through real-world experiments on a mobile robot using a stereo camera to highlight its robustness and viability for real robotic applications. Experimental results demonstrate that the proposed system outperforms the original ORB-SLAM3, reducing the error by 93% in the public TUM dataset while preserving computational efficiency.


  • Open Access

    Comparative Analysis of Convolutional Neural Networks for Optical Fiber Classification

    by Carlos Cezar de Oliveira Conrad, Marcelo Eugenio Manfrim Mafalda, Davi Klein, Rafael Crespo Izquierdo, Gabriel Alejandro Tarnowski, Guillermo Mudry, Anselmo Rafael Cukla, Daniel Fernando Tello Gamarra

    This paper employs deep learning algorithms to address the classification of optical fiber cables through image processing techniques. The VGG-16 and Res-Net50 neural network architectures are utilized, and their performance is evaluated in comparison. The experiments were developed in an optical fiber dataset composed of optical fiber images and increased with synthetic data using the data augmentation technique; the dataset has optical fiber images in different states of preservation and colors. The neural networks will classify the images into two classes ("good" or "bad").