Search
Now showing items 1-9 of 9
Communities or Collections matching your query
Small scale and headwater catchments are mostly ungauged, even though their observation could help to improve the understanding of hydrological processes. However, it is expensive to build and maintain conventional measurement networks. Thus, the heterogeneous characteristics and behavior of catchments are currently not fully observed. This study introduces a method to capture water stage with a flexible low-cost camera setup. By considering the temporal signature of the water surface, water lines are automatically retrieved via image processing. The image coordinates are projected into object space to estimate the actual water stage. This requires high resolution 3D models of the river bed and bank area which are calculated in a local coordinate system with SfM, employing terrestrial as well as UAV imagery. A medium- and a small-scale catchment are investigated to assess the accuracy and reliability of the introduced method. Results reveal that the average deviation between the water stages measured with the camera gauge and a reference gauge are below 6 mm in the medium-scale catchment. Trends of water stage changes are captured reliably in both catchments. The developed approach uses a low-cost camera design in combination with image-based water level measurements and high-resolution topography from SfM. In future, adding tracking algorithms can help to densify existing gauging networks.
In this paper an automatic approach is proposed to measure flow velocity with an uncooled thermal camera. Hot water is used as thermal tracer. The introduced tracking algorithm utilizes the pyramidal Lucas-Kanade method and is especially suitable for thermal image data. The performance of the new tool is compared to traditional image-based tracking tools, i.e. PIVlab and PTVlab. Experiments are performed in the laboratory for three different flow velocities. Afterwards, tests are conducted in a small stream to illustrate the suitability of the tool for field measurements. Results of the laboratory experiments as well as of the field experiments show that our tracking algorithm, applied to imagery from a thermal camera, outperforms commonly used tracking methods. Our tool provides velocity fields with very high resolution and is in close agreement with reference measurements, whereas PTVlab and PIVlab tend to overestimate and underestimate flow velocities, respectively.
An automatic workflow is introduced, including an image-based tracking tool, to measure surface flow velocities in rivers. The method is based on PTV and comprises an automatic definition of the search area for particles to track. Tracking is performed in the original images. Only the final tracks are geo-referenced, intersecting the image observations with water surface in object space. Detected particles and corresponding feature tracks are filtered considering particle and flow characteristics to mitigate the impact of sun glare and outliers. The method can be applied to different perspectives, including terrestrial and aerial (i.e. UAV) imagery. To account for camera movements images can be co-registered in an automatic approach. In addition to velocity estimates, discharge is calculated using the surface velocities and wetted cross-section derived from surface models computed with SfM and multi-media photogrammetry. The workflow is tested at two river reaches (paved and natural) in Germany. Reference data is provided by ADCP measurements. At the paved river reach highest deviations of flow velocity and discharge reach 5% and 4%, respectively. At the natural river deviations are larger (26% and 20%, respectively) due to the irregular cross-section shapes hindering accurate contrasting of ADCP- and image-based results. The provided tool enables the measurement of surface flow velocities independently of the perspective from which images are acquired. With the contact-less measurement spatially distributed velocity fields can be estimated and river discharge in previously ungauged and unmeasured regions can be calculated.
A workflow is introduced to automatically measure water stages based on image measurements using deep learning. So far, most camera gauges do not provide the needed robustness to achieve accurate water stage measurements because of changing environmental conditions. The novel, suggested approach is based on two CNNs (i.e. FCN and SegNet) to identify water in imagery. The image information is transformed into metric water level values intersecting the extracted water contour with a 3D model. The workflow allows for the densification of river monitoring networks based on low-cost camera gauges in various scenarios.
This repository contains a dataset from a calibration campaign of 20 low-cost rain gauges, data of 3 reference gauges and a dataset of the labcalibration of 66 low-cost raingauges. Further, source code for Arduino and Raspberry Pi based Dataloggers is provided.