Methodology_Type: Field
Methodology_Description:
We flew a small unoccupied aerial system (sUAS; Solo 3D Robotics, Berkeley, CA) over walruses hauled out on shore near Point Lay, Alaska. We fit the sUAS with a 41 degree angle gimbal-mounted camera (Peaupro41; Peau Productions, San Diego, CA) and flew transects over walrus haulouts at 107 to 112 m altitude using autonomous flight commands (Tower app:
https://github.com/DroidPlanner/Tower ). We georeferenced aerial images with coordinates estimated on-board the sUAS at approximately 1.2 second intervals during image collection based on a Global Navigation Satellite System (GNSS) which received and processed signals from both the U.S. Global Navigation System constellation (GPS) and the Russian global navigation system (GLONASS; Russian: ГЛОНАСС, Глобальная навигационная спутниковая система). Location was interpolated using input from a printed circuit board inertial measurement unit (IMU). Coordinates were recorded in the WGS84 coordinate reference system (EPSG:7660), with elevation specified in meters relative to the WGS84 ellipsoid. We assigned approximate image acquisition coordinates to the image metadata from the image timestamp and locations collected on-board the sUAS using the GeoSetter program (v.3.5.0, www.geosetter.de, Friedmann Schmidt) which is based on exiftools (ver. 12.21,
https://exiftool.org, Phil Harvey). Because we have no survey-grade ground control points for cross-validation and calibration of the GNSS locations collected on-board the UAS, these locations are provided without a specified location accuracy report. Surveys were flown under FAA part 107 rules that stipulated visual line of sight be maintained by the remote pilot or by a trained visual observer throughout the UAS flight. All surveys were conducted under Marine Mammals Protection Act permit number MA801652-7 at altitudes and in a manner that minimized potential disturbances to walruses resting on land. Although it was not possible to conduct independent observations of walruses during the surveys, inspection of the survey imagery and field observers detected no changes in walrus behavior, such as head raises or displacements, in apparent response to sUAS overflights. Access to the lands was arranged through the landowners and authorized under the North Slope Borough permit 18-449.
Process_Description:
We generated orthoimages with approximately 1.5 to 2.2 cm/pixel resolution from the georeferenced images using structure from motion software (Agisoft version 1.5.5, Saint Petersburg, Russia). We aligned the images, performed an initial bundle adjustment for the following parameters: f, cx, cy, k1, k2, k3, p1, p2, then performed gradual selection to systematically reduce spatial error in the sparse cloud. For the gradual selection, we iteratively culled sparse points based on a reconstruction uncertainty threshold of 10 and followed by a bundled adjustment using the same parameters as before. We then culled sparse points that exceeded the projection accuracy threshold of 3, performing a bundle adjustment as before. We then culled sparse points that exceeded a re-projection error threshold of .3, performing a bundle adjustment on all parameters (f, cx, cy, k1, k2, k3, b1, b2, p1, p, p3, p4). We iterated each of these gradual selection processes until the specified parameter level was attained using the python API for the Agisoft software if the images were of superior quality. If survey images suffered image blur or erratic camera angles due to gusty winds, we relaxed error reduction constraints by using a reconstruction uncertainty threshold of 12, a projection accuracy threshold of 3, and a reconstruction uncertainty threshold of 4, and limited iterations to not cull more than 50% of the sparse points during the reconstruction uncertainty and projection accuracy culling and not allow culling of more than 10% of the points during the re-projection accuracy culling, so as to enable construction of a structure from motion model across the walrus haulout. We then built a dense cloud, using a medium depth filtering and built a medium density height field mesh with interpolation enable. From this mesh we built the orthoimages in a WGS84 coordinate system from a mosaic of images overlain on the mesh. The U.S. Geological Survey, Advanced Research Computing Center and the Pacific Coastal and Marine Science Center supported this processing.
Process_Date: Unknown