Banca de QUALIFICAÇÃO: ARTHUR ANDRADE BEZERRA

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : ARTHUR ANDRADE BEZERRA
DATE: 20/12/2025
TIME: 13:00
LOCAL: Sala Virtual PPgEEC
TITLE:

Enhancing Visual Navigation Techniques for an Anti-Collision System in an Autonomous Robotic Sailboat

KEY WORDS:

Robotics Vision; Aquatic Visual Dataser; Classification

PAGES: 55
BIG AREA: Ciências Exatas e da Terra
AREA: Ciência da Computação
SUBÁREA: Metodologia e Técnicas da Computação
SPECIALTY: Processamento Gráfico (Graphics)
SUMMARY:

Autonomous navigation in maritime environments is still relatively unexplored compared to advancements in land and air vehicles, despite its potential for continuous environmental monitoring, support for scientific research, and increased safety in long-duration missions. In this context, this thesis investigates the use of computer vision in the perception and avoidance of collisions in autonomous robotic sailboats, using the N-Boat and F-Boat, developed at Brazilian universities, as reference platforms. The work starts from a classic, entirely geometric baseline that combines stereo calibration and rectification, color-based sky-sea segmentation, disparity estimation by StereoSGBM, and fusion with inertial data in a polar visual sonar. Subsequently, a neural perception architecture is proposed that integrates semantic water segmentation, obstacle detection with modern detection models (including YOLOv11 and RF-DETR), and distance estimation, enhanced by the SAHI technique to improve the detection of very small objects in high-resolution images. Due to the scarcity of annotated data in maritime scenarios, the VISBOAT dataset and a programmatic augmentation module were developed, responsible for generating photometrically coherent synthetic images from real scenes. Preliminary experiments indicate consistent performance gains in detection metrics (mAP@50), especially for buoys and small distant obstacles, when combining real data, generative augmentation, and SAHI. These results support the hypothesis that a visual perception architecture, supported by stereo vision, segmentation, and object detection, can provide sufficient information to support navigation and collision avoidance systems in autonomous sailboats.

COMMITTEE MEMBERS:
Presidente - 1345674 - LUIZ MARCOS GARCIA GONCALVES
Externo à Instituição - COSIMO DISTANTE - US
Externo à Instituição - ESTEBAN WALTER GONZALEZ CLUA - UFF