Fusion model of unmanned aerial vehicle-based digital elevation model for accuracy improvement mapping

Most construction projects require accurate Digital Elevation Model (DEM) data adhere to a certain standard of accuracy both in the horizontal and vertical components. However, due to the DEM inherent errors, it has been a major research concern to generate accurate DEM over large and inaccessible areas in a cost and time effective manner. Hence, the fusion of UAV-based DEMs. This study aimed: (i) to develop models for investigating the effect of Atmospheric Pressure (AP) on DEMs generated from fixed-wing UAV platform, (ii) to develop a methodological approach for determining an optimum flying altitude for UAV cadastral mapping and (iii) to develop fusion and filtering algorithms for improving the vertical accuracy of DEM generated by UAV systems. Before aerial photography, forty-five ground control points (GCPs) were established evenly in the study area using a real-time kinematic differential global positioning system for georeferencing and quality assessment of UAV products (DEM and orthoimage). Regarding the first objective, a canon digital camera onboard fixed-wing UAV was flown over UniPutra golf club in the Universiti Putra Malaysia campus at an altitude of 100 m, 150 m, 200 m, 250 m, 350 m, 400 m, and 500 m. The onboard camera took a series of overlapping photographs of the study area at a predefined three seconds regular time interval. The photos were processed using Agisoft algorithm. In the end, seven DEMs were exported in tiff file format. The DEMs were evaluated for atmospheric pressure effect using a proposed mathematical model. The results of the AP effect on the DEM at 100 m, 150 m, 200 m, 250 m, 350 m, 400 m, and 500 m altitudes produced 0.072 m, 0.05 m, 0.014 m, 0.01m, 0.004 m, 0.003 m, and 0.002 m, respectively. The results were verified using the height of established GCPs and corresponding points on the DEMs. The verification process provides RMSE of 0.03 m, 0.05 m, 0.07 m, 0.1 m, 0.13 m, 0.14 m, and 0.16 m, respectively. The results show that the DEM produced at an altitude of 100 m generated a higher accuracy of 0.03 m despite a huge AP effect of 0.072 m. Conversely, at 500 m altitude, a lesser AP effect of 0.02 produced a low-quality DEM of 0.16 m. Analysis of variance (ANOVA) test conducted to uncover the interacting effects of AP on the DEMs produced 0.931 (R-value), 0.867 (R Square), and 0.017 (significance F) correlation coefficients. The results of the test indicated goodness of fit because both Multiple R and R Square values are very close to 1 and Significance F value < 0.05. The overall results of the experiment show that the effect of AP is insignificant and, thus, can be ignored. To achieve the second objective, Tarot 680- hexacopter UAV was flown over the stadium UPM near the UniPutra golf club at the Universiti to take photographs at an altitude range of 70 m, 100 m, and 250 m. The photographs were processed and georeferenced using an Agisoft PhotoScan algorithm and ten of established ground control points, respectively. The resulting orthoimages exported to an ArcGIS software for cadastral map digitization. Analyses, such as visual, tabular, and graphical, were carried out to examine an optimum flight altitude for cadastral mapping. The results of the study show that the cadastral map at 70 m altitude produced an optimum result. The last objective proposes a fusion model that integrates a weighted averaging and median additive filtering to improve the quality of DEM derived from fixed-wing UAVs. The fixed-wing DEM was fused with highquality DEM generated from a multi-rotor UAV platform. Assessment of the DEM produced root mean square error of 1.14 cm and standard vertical accuracy of 2.24 cm at a 95% confidence level (CL). This value represents a decrease in the vertical standard error of 18.31 cm to 2.24 cm, which is an improvement of 87.77%. The result of the study indicated that the method is suitable for improving a low-quality DEM produced by the UAV system. Transferability assessment of the proposed model was conducted by fusing a high-accuracy DEM generated from a LiDAR dataset (1.87 cm) with a low-accuracy fixed-wing DEM. The resulting DEM was filtered for noise reduction. Accuracy assessment of fused and filtered DEM at 95% CL produced 2.13 cm, which is a gap of 16.18 cm (88.37%) when compared with a low-accuracy input DEM. The results show that the method is not only efficient for improving UAV DEM quality but also can be used to improve the quality of DEM derived from other sensors.

Text FK 2020 49 ir.pdf Download (1MB)

Actions (login required)

View Item