Structure-from-Motion (SfM) surface models derived from seafloor video from the Channel Islands, California

Data Collection: Video was collected during USGS field activity 2016-668-FA in the Channel Islands, CA. Video footage was collected using the USGS camera sled, which was equipped with two high resolution video cameras, one pointing downward (vertical) and the other pointing slightly forward (oblique) and was towed by the research vessel Shearwater. The camera sled was also equipped with paired lasers to provide scaling to features on the seafloor. Video was recorded in depths approximately 10 to 70 meters. Shipboard GPS locations were recorded every 1 second into one of the audio channels of the video using Red Hen Systems, LLC hardware.

Structure-from-Motion Processing: Still frames were extracted from the video at a constant rate of 1 frame every 2/3 second. For the shipboard GPS method 6 markers were selected throughout the video imagery. Those markers were assigned latitude and longitude values from the shipboard GPS according to the playback time of that video frame, and the markers were assigned the depth from the existing sonar that corresponded to that location. For the georeferencing method, features visible in the video imagery were georeferenced to that feature where it was visible in the sonar data (7 markers were assigned this way). A lens model was developed using the BOBSled cameras to record imagery of a calibration board in a salt-water tank. That imagery was used to create a model, and the SfM software's lens model was exported as the fixed lens model. Automatic calibration by the SfM software was also used. Four models were produced: calibrated using a fixed lens model with the shipboard GPS that was embedded in the audio ("SfM_Cal-ShipGPS_UTM10N.tif"), lens model auto-calibrated by the SfM software with ship GPS ("SfM_AutoCal-ShipGPS_UTM10N.tif"), calibrated using a fixed lens model with locations derived from georeferencing the video imagery to the existing sonar bathymetry ("SfM_Cal-Georef_UTM10N.tif"), and auto-calibrated with georeferenced locations ("SfM_AutoCal-Georef_UTM10N.tif"). SfM models were created using Agisoft PhotoScan version 1.3.5. Photos were aligned in PhotoScan, iteratively 'cleaned' to reduce error using the following steps in PhotosScan: all points with reconstruction uncertainty greater than 10 were deleted (iteratively - no more than 50 percent of the models points were deleted at a time); all points with projection accuracy greater than 3 were deleted; reprojection error was reduced to 0.3 by iteratively deleting no more than 10 percent of the models points; after each of the previous steps, cameras were optimized; a dense point cloud was created in PhotoScan; digital elevation model created from dense point cloud in PhotoScan; elevation model exported as TIF image.

Edited metadata to add keywords section with USGS persistent identifier as theme keyword. No data were changed.