Remoção automática de objetivos indesejados em nuvens de pontos utilizando o algoritmo de detecção votenet

Abstract

This work explores the automatic removal of unwanted objects in point clouds, a complex but essential task in the field of computing applied to civil engineering and architecture. Point clouds are three-dimensional representations generated by technologies such as laser scanners and photogrammetry, widely used in the reconstruction of BIM (Building Information Modeling) representations and in the digitization of buildings. However, these clouds often contain irrelevant objects that impair the segmentation of architectural and structural elements. The objective of this study was to implement the VoteNet algorithm, an object detection technique based on deep learning, to automatically identify and remove these objects. To evaluate the effectiveness of the method, metrics such as the Jaccard Index and Hausdorff Distance were used, which allowed measuring the precision and accuracy of the removal process. The results indicate that VoteNet performed well in detecting and removing unwanted objects, significantly improving the quality and usability of point clouds for future segmentation and modeling steps. The conclusion highlights the potential of the algorithm to optimize the automation workflow in civil construction, making data processing more efficient and robust.