Photogrammetric models have become a standard tool for the study of surfaces, structures and natural elements. As an alternative to Light Detection and Ranging (LiDAR), photogrammetry allows 3D point clouds to be obtained at a much lower cost. This paper presents an enhanced workflow for image-based 3D reconstruction of high-resolution models designed to work with fixed time-lapse camera systems, based on multi-epoch multi-images (MEMI) to exploit redundancy. This workflow is part of a fully automatic working setup that includes all steps: from capturing the images to obtaining clusters from change detection. The workflow is capable of obtaining photogrammetric models with a higher quality than the classic Structure from Motion (SfM) time-lapse photogrammetry workflow. The MEMI workflow reduced the error up to a factor of 2 when compared to the previous approach, allowing for M3C2 standard deviation of 1.5 cm. In terms of absolute accuracy, using LiDAR data as a reference, our proposed method is 20% more accurate than models obtained with the classic workflow. The automation of the method as well as the improvement of the quality of the 3D reconstructed models enables accurate 4D photogrammetric analysis in near-real time.
20 October 2023 in Natural Hazards and Earth System Sciences
In this publication we address the lack of technical expertise in the geoscience community in the design and construction of photogrammetric systems for monitoring natural hazards at high spatio-temporal resolution. Accordingly, we provide in-depth information on the components, assembly instructions, and programming codes required to build them, making them accessible to researchers from different disciplines who are interested in 3D change detection monitoring. Each system comprises five photographic modules and a wireless transmission system for real-time image transfer. As an alternative to lidar (light detection and ranging), high-end digital cameras offer a simpler and more cost-effective solution for the generation of 3D models, especially in fixed time-lapse monitoring systems. The acquired images, in combination with algorithms that allow the creation of improved 3D models, offer change detection performance comparable to lidar. We showcase the usefulness of our approach by presenting real-world applications in the field of geohazard monitoring. Our findings highlight the potential of our method to detect pre-failure deformation and identify rockfalls with a theoretical change detection threshold of only 3–4 cm, thereby demonstrating the potential to achieve similar accuracies to lidar but at a much lower cost. Furthermore, thanks to the higher data acquisition frequency, the results show how the overlap of events that leads to an erroneous interpretation of the behaviour of the active area is minimized, allowing, for example, more accurate correlations between weather conditions and rockfall activity.
The emerging use of photogrammetric point clouds in three-dimensional (3D) monitoring processes has revealed some constraints with respect to the use of LiDAR point clouds. Oftentimes, point clouds (PC) obtained by time-lapse photogrammetry have lower density and precision, especially when Ground Control Points (GCPs) are not available or the camera system cannot be properly calibrated. This paper presents a new workflow called Point Cloud Stacking (PCStacking) that overcomes these restrictions by making the most of the iterative solutions in both camera position estimation and internal calibration parameters that are obtained during bundle adjustment. The basic principle of the stacking algorithm is straightforward: it computes the median of the Z coordinates of each point for multiple photogrammetric models to give a resulting PC with a greater precision than any of the individual PC. The different models are reconstructed from images taken simultaneously from, at least, five points of view, reducing the systematic errors associated with the photogrammetric reconstruction workflow. The algorithm was tested using both a synthetic point cloud and a real 3D dataset from a rock cliff. The synthetic data were created using mathematical functions that attempt to emulate the photogrammetric models. Real data were obtained by very low-cost photogrammetric systems specially developed for this experiment. Resulting point clouds were improved when applying the algorithm in synthetic and real experiments, e.g., 25th and 75th error percentiles were reduced from 3.2 cm to 1.4 cm in synthetic tests and from 1.5 cm to 0.5 cm in real conditions.
11 December 2024 in IEEE Transactions on Geoscience and Remote Sensing
In this study, we present an approach to exploit the keypoint R-convolutional neural network (CNN) structure for automated detection of ground control points (GCPs) in images acquired by optical sensors. Our deep learning methodology employs fine-tuning on three distinct datasets, thereby enabling the specific addressing of inherent variations in GCP types, including shape, size, and color. Our approach uses an end-to-end artificial intelligence (AI)-based approach that does not require any postprocessing of the data or the use of specific GCPs. Performance metrics are evaluated to compare the AI results with manually labeled data, coordinates obtained by an ellipse fitting algorithm for circular targets, and a semi-automatic optical flow algorithm for arbitrary targets (e.g., crosses). The study highlights the importance of a dataset-specific fine-tuning and data augmentation strategy to enhance the model’s accuracy in locating GCPs. The results demonstrate the effectiveness of the R-CNN keypoint approach in successfully identifying GCPs with subpixel accuracy compared to manual labeling in all datasets. Our AI-based approach outperformed semi-automatic methods based on computer vision algorithms, identifying GCPs in images taken under adverse conditions and in cases where GCPs were partially obscured. Furthermore, the results demonstrate robust performance in a transferability test, that is, identifying GCPs in different images obtained at different locations and with different cameras than those used for training.
12 February 2026 in Hydrology and Earth System Sciences
The study presents a robust, automated camera gauge for long-term river water level monitoring operating in near real-time. The system employs artificial intelligence (AI) for the image-based segmentation of water bodies and the identification of ground control points (GCPs), combined with photogrammetric techniques, to determine water levels from surveillance camera data acquired every 15 min. The method was tested at four locations over a period of more than 2.5 years. During this period almost 218 000 images were processed. The results demonstrate a high performance, with mean absolute errors ranging from 0.96 to 2.66 cm in comparison to official gauge references. The camera gauge demonstrates resilience to adverse weather and lighting conditions, achieving an image utilisation rate of above 95 % throughout the entire period. The integration of infrared illumination enabled 24/7 monitoring capabilities. Key factors influencing absolute error were identified as camera calibration, GCP stability, and vegetation changes. The low-cost, non-invasive approach advances hydrological monitoring capabilities, particularly for flood detection and mitigation in ungauged or remote areas, enhancing image-based techniques for robust, long-term environmental monitoring with frequent, near real-time updates.
Between June and August 2022, the European Forest Fire Information System (EFFIS) reported more fires in Europe than in any other recent summer season. This is particularly true for Central Europe, where the largest forest fire in recent Czech history occurred in the German-Czech border region. With global warming and resulting longer dry periods, the length and severity of wildfire seasons in central Europe will likely increase. Therefore, easy to implement and cost-effective methods to assess wildfire damage and regeneration of the ecosystems are getting increasingly important. In this study we evaluated how different datasets obtained by uncrewed aerial system (UAS) can be incorporated with datasets obtained from the ground to describe the fire affected landscape. Thereby, multi-spectral 3D point clouds were derived from low-cost UAV laser scanning and using structure from motion (SfM) photogrammetry applied to RGB and multi-spectral imagery. The aerial datasets were combined with ground-based terrestrial and mobile laser scanning. The datasets were acquired in several surveys following the forest fire event in the German part of the National park Saxonian/Bohemian Switzerland.Initial results show the potential of UAS-based sensing for efficient mapping of a burned area with a high resolution (600-1000 pts/m²). The combination of point clouds from UAS-based laser scanning and photogrammetry enables a detailed representation of the burned forest with different levels of fire damage (e.g., in still present canopy) when compared to the single datasets. The UAS based laser scanning data reveals higher noise compared to the SfM-based point clouds. However, the accuracy is still sufficient to improve the quality of orthomosaics in densely vegetated areas. In a next step, further investigations on data accuracy are conducted and automated point cloud fusion algorithms based on classified points are considered.