Investigating the accuracy of Digital Elevation Models from UAV images in areas with low contrast : a sandy beach as a case study

English. Elevation models are important in many applications in Geographical Information System (GIS), both when applying environmental analyses and in urban planning studies. The production of elevation models has been a difficult and expensive task using traditional surveying techniques. Remote sensing methods for producing digital elevation models have solved the difficult part of the task, which consist of collecting high resolution elevation data; however, the cost of elevation datasets has resulted in the fact that they have been among the least updated GIS datasets. Using Unmanned Aerial Vehicles (UAVs) is an effective and inexpensive approach for data collection that can facilitate the production of high temporal and spatial resolution elevation models. The method used for obtaining 3D models from overlapping images has been known for decades. Software that uses scale invariant object recognition methods to obtain 3D models from multiple images has been used both in computer vision and GIS applications. This study investigates the quality of elevation models obtained from images in areas with low elevation changes and low image contrast, i.e. coastal areas. The influence of image pixel size on the quality of the results is studied, as well as how the number and distribution of the Ground Control Points (GCPs) influence the accuracy of the models. The outputs of the study show a high correlation between pixel size and the quality of the obtained elevation models. While the number and distribution of the GCPs have a strong impact on the results, neither number of overlapping images nor different land cover types record a clear effect on the quality of the calculated GCPs. The difference between a number of elevation models express a systematic distribution of error with lower error values around the GCPs. The study demonstrates that time and resources can be saved and still obtain better results by selecting the best flight height with the optimal number and distribution of the GCPs.

English. Elevation data is important in many applications and studies, both in environmental analyses and in urban planning. Production of elevation data has always been difficult and expensive. Remote sensing has solved the difficult part of the task, which is collecting high resolution elevation data; however, the cost of these datasets made them still one of the least updated GIS datasets. Using drones is an effective and inexpensive approach for production of detailed elevation models frequently. The method used for obtaining 3D models from overlapping images has been known for decades. And the software that uses object recognition methods to obtain 3D models from images has been used both in computer vision and GIS applications. This study investigates the quality of elevation models obtained from images in areas with low elevation changes and low image contrast, i.e. coastal areas. The influence of image pixel size on the quality of the results is studied, as well as how the number and distribution of the Ground Control Points (GCPs) influence the accuracy of the models. The outputs of the study confirm the relation between pixel size and the quality of the results. The results demonstrate a strong effect of the number and distribution of the GCPs on the quality of the results. However, number of overlapping images and different land cover types do not record a clear effect. The study shows that time and resources can be saved and still obtain better results by selecting the best flight height with the optimal number and distribution of the GCPs.