Avaliação da Acurácia da Nuvem de Pontos Gerada por Processamento Fotogramétrico de Imagens Adquiridas por Sensores Embarcados em ARP para Levantamentos Topográficos de Alta Precisão

Abstract

The advancement of geotechnologies and the consolidation of Remotely Piloted Aircraft (RPA) have significantly expanded the use of digital photogrammetry in the generation of high-resolution cartographic products, especially for topographic surveys and agricultural applications. In this context, this dissertation aimed to evaluate the three-dimensional accuracy of the point cloud generated through photogrammetric processing of images acquired by sensors onboard an RPA, using observations obtained with a Global Navigation Satellite System (GNSS) receiver operating in Real Time Kinematic (RTK) mode as reference. The study area corresponds to a section of Highway MG-223, between kilometers 36 and 39, in the municipality of Estrela do Sul, Minas Gerais, Brazil, characterized by rugged terrain and significant altimetric variation, totaling approximately 60 hectares. A total of 44 ground control points were established, including 26 Ground Control Points (GCPs) and 18 Check Points (CKPs), whose coordinates were determined using a dual-frequency GNSS receiver and referenced to the RBMC MGMT station in Monte Carmelo, Minas Gerais. The aerial photogrammetric survey was conducted following an appropriate flight plan, and the images were processed using the Structure from Motion (SfM) and Multi-View Stereo (MVS) algorithms in Agisoft Metashape, resulting in a dense and georeferenced point cloud. External validation of the model was performed by comparing 2,350 GNSS-RTK points with the corresponding coordinates extracted from the point cloud, analyzing discrepancies in the East E(X), North N(Y), and Height h(Z) components, as well as statistical metrics such as RMSE, MAE, and maximum and minimum errors. The results indicated RMSE values below 5 cm for the horizontal components and below 10 cm for the vertical component, classifying the generated product as Class A according to the Brazilian Positional Accuracy Standard (PEC/EP) and Level 1 (highest accuracy) according to ASPRS (2014). It is concluded that RPA-based photogrammetry, when combined with rigorous geodetic control and appropriate methodological procedures, is capable of producing point clouds with high metric reliability, proving to be a technically viable and efficient alternative for high-precision topographic surveys and applications in agriculture and roadway infrastructure.