Applying photogrammetry to time-lapse imagery for geomorphological change detection

Cameras that capture images in time-lapse mode of the earth surface enable great opportunities for change detection and thus potential process identification and understanding. The camera systems can range from simple and robust game cameras to complex and synchronised full frame cameras. The main workflow of calculating digital elevation models from overlapping images is similar for the different types of systems; automatically matching the images, performing bundle adjustment considering either calibrated or non-calibrated cameras, geo-referencing the data by automatic ground control point (GCP) measurement, densifying the point cloud and eventually calculating point cloud differences. However, adapted pre-and post-processing steps are needed due to the varying observation conditions considering the camera qualities and the objects of interest. The time-series of point cloud-based change information can be further processed, for example, with time-series clustering approaches to disentangle overlapping processes.

We will introduce three different case studies in the field of fluvial geomorphology, soil erosion research and rockfall assessment. Thereby, different camera systems are utilized. Four low-cost time-lapse cameras are applied in arctic environments to study changes of a river bank at a distance of about 60 m. The high robustness of the cameras encompasses the trade-off of low quality images. In addition, challenging lighting conditions and enduring snow cover complicate the photogrammetric processing. The images are captured with a frequency of two hours, and six permanent GCPs are used to geo-reference the measurements.

Digital SLR cameras are used in moderate climate to measure soil surface changes either due to rainfall simulations or due to natural rainfall events. During the rainfall simulation we use images that are captured by up to ten cameras with a frequency of 10 to 20 seconds and at an object distance between 3 to 4 m. And at the field plot we installed three special camera rigs that encompass five cameras each that are event-controlled by a micro-controller and single board computer solution, which trigger the cameras each time a rain collector bucket is tipping in addition to daily captured images. Challenges for change detection arise from vegetation present at the plots and from runoff water covering the soil surface. Eventually, the derived models of change are used to validate physical based soil erosion models.

The last case study utilizes five full-frame system cameras in the Mediterranean to detect single rockfall events. Images are captured three times a day by an ad-hoc system at a distance of about 100 m. The data is transferred via a locally installed network module. Many areas within the field of view remain stable throughout the measurement period allowing for a time-SIFT approach that matches the images from different points in time. Machine learning algorithms are applied to automatically identify rockfalls in the final 4D dataset. Thereby, we showcase the great potential of time-lapse photogrammetry for different applications of geomorphological change detection.