Akademik Veri Yönetim Sistemi
31st International Conference on Artificial Life and Robotics, ICAROB 2026, Oita, Japonya, 29 Ocak - 01 Şubat 2026, ss.139-144, (Tam Metin Bildiri)
Urban trees play a crucial role in enhancing city ecosystems and urban aesthetic value; however, their maintenance, particularly through pruning, is essential for infrastructure safety. Traditional manual pruning is labor-intensive, timeconsuming, and poses risks to workers. To overcome these limitations, automated approaches are becoming increasingly important. Automation in urban tree pruning typically begins with the visual perception of tree structures. However, many existing tree vision systems rely on 3D scanning techniques, which, although accurate, require substantial computational resources to operate. This study introduces a semi-automated, rule-based 2D vision pruning decision approach designed for low-complexity and minimally occluded urban tree structures. By utilizing YOLOOBB, a deep learning model trained on a diverse set of urban tree images, the proposed method accurately detects branch positions from a single RGB view using oriented bounding boxes. Pruning decisions are then determined using explicit, operator-defined height rules, where branches are selected for crown raising or crown reduction solely based on their relative vertical positions with respect to predefined safety thresholds. These detections serve as the basis for identifying and selecting branches for either raising or reduction. The experimental results demonstrate detection accuracy exceeding 90% for both branches and trunk, and pruning simulations show consistent and controllable crown raising and reduction based on height thresholds. This approach provides a scalable, low-computation alternative to 3D-based systems for managing urban trees, ensuring safety and ecological benefits in urban environments while preserving human control over pruning decisions.