Visão computacional para estimativa de profundidade e detecção de linhas e estruturas em redes de energia elétrica
Saved in:
| Hovedforfatter: | |
|---|---|
| Publication Date: | 2026 |
| Format: | Master thesis |
| Sprog: | por |
| Source: | Repositório Institucional da UPF |
| Download full: | https://repositorio.upf.br/handle/123456789/10273 |
Summary: | Vegetation encroachment within the rights-of-way of power transmission and distribution lines poses a significant risk to the safety and reliability of electrical systems, making periodic pruning, felling, and maintenance operations indispensable. Accurate, real-time detection of vegetation and nearby electrical structures is essential to preventing incidents such as short circuits, fires, and power supply interruptions. This paper introduces Depth Line, a modular and scalable computer vision system designed to support both local and serverside inference, and is adaptable across multiple platforms. The system enables the real-time detection of power lines and associated structures, as well as accurate distance estimation to assist mechanical pruning operations. The proposed approach considers RGB images, applying the YOLO11 architecture for semantic segmentation of objects of interest and the Depth Anything v2 model for monocular depth estimation. This eliminates the need for additional sensors such as LiDAR or stereo systems and reduces overall system costs. Experiments were conducted under computational resource constraints, simulating conditions similar to those of embedded systems. The results indicated consistent performance of the segmentation model, with Dice ≈ 0.66 and mAP@0.5 ≈ 0.65. The depth estimation model presented greater stability at intermediate distances; however, significant metric errors were observed at short distances, with RMSE ≈ 3.5 meters and MAE ≈ 2.6 meters. These findings demonstrate the feasibility of integrating semantic segmentation and real-time depth estimation as a support tool for operational safety, while also highlighting limitations that point to the need for metric calibration and further model refinement in future work. |
Lignende værker: Visão computacional para estimativa de profundidade e detecção de linhas e estruturas em redes de energia elétrica
- Estratégias de otimização de trajetos e alocação de torres em projetos de linhas de transmissão aéreas
- Cálculo de parâmetros de malhas de aterramento em geometrias não convencionais
- Proposta de implementação de melhorias no processo de execução das obras de distribuição de energia elétrica : aplicado a uma empresa prestadora de serviços
- BIM aplicado a obras de infraestrutura (BiiM) : mini e pequenas centrais hidrelétricas
- Cálculo de campos elétricos e magnéticos nas proximidades de linhas de transmissão: uma abordagem analítica e numérica
- Ambiente virtual para reabilitação de membros superiores utilizando visão computacional