In forestry, aerial photogrammetry by means of Unmanned Aerial Systems (UAS) could bridge the gap between detailed fieldwork and broad-range satellite imagery-based analysis. However, optical sensors are only poorly capable of penetrating the tree canopy, causing raw image-based point clouds unable to reliably collect and classify ground points in woodlands, which is essential for further data processing. In this work, we propose a novel method to overcome this issue and generate accurate a Digital Terrain Model (DTM) in forested environments by processing the point cloud. We also developed a highly realistic custom simulator that allows controlled experimentation with repeatability guaranteed. With this tool, we performed an exhaustive evaluation of the survey and sensor settings and their impact on the 3D reconstruction. Overall, we found that a high frontal overlap (95%), a nadir camera angle (90°), and low flight altitudes (less than 100 m) results in the best configuration for forest environments. We validated the presented method for DTM generation in a simulated and real-world survey missions with both fixed-wing and multicopter UAS, showing how the problem of structural forest parameters estimation can be better addressed. Finally, we applied our method for automatic detection of selective logging.
22 June 2013 in Journal of Real-Time Image Processing
En este trabajo se presenta un método probabilístico para resolver el problema de la localización de un robot diferencial. Se usa el Filtro Extendido de Kalman (EKF) para fusionar la información obtenida por registraciones de mediciones láser mediante ICP (IterativeClosest Point) con la información de odometría provista por encoders. Para utilizar EKF es necesario estimar la covarianza de cada fuente de información, sin embargo el algoritmo ICP no devuelve la covarianza asociada. En este artículo se describe una forma de calcular esta covarianza. Los resultados obtenidos muestran que el método de fusión de sensores resulta en una estimación más precisa de la pose del robot en comparación con las estimaciones que se podrían obtener mediante odometría e ICP individualmente.
A dense SLAM system is essential for mobile robots, as it provides localization and allows navigation, path planning, obstacle avoidance, and decision-making in unstructured environments. Due to increasing computational demands the use of GPUs in dense SLAM is expanding. In this work, we present coVoxSLAM, a novel GPU-accelerated volumetric SLAM system that takes full advantage of the parallel processing power of the GPU to build globally consistent maps even in large-scale environments. It was deployed on different platforms (discrete and embedded GPU) and compared with the state of the art. The results obtained using public datasets show that coVoxSLAM delivers a significant performance improvement considering execution times while maintaining accurate localization. The presented system is available as open-source on GitHub https://github.com/lrse-uba/coVoxSLAM.