Laboratory Observations of Variable Size and Shape Particles-Artificial Sand and Oil Agglomerates: June 2017 Video Data

In June 2017, experiments were conducted at NRL's Sediment Dynamics Laboratory, located in Stennis, MS. Particles with spherical, patty, ellipsoidal, and angular-ellipsoidal shapes representing aSOAs were deployed in the small-oscillatory flow tunnel with a cross-sectional area of 25 cm by 25 cm and a test bed length of 2-meters (m). Oscillating currents (simulating bottom velocities generated by waves) were driven in the tank by a flywheel with variable frequencies (5-90 rotations per minute) and stroke lengths (22, 33, or 44 cm). Five cameras (two Canon 7D DSLR and three GoPro Hero3+ were mounted outside the tank to record particle movement, in HD, at 29.97 fps and 1920x1080 pixel resolution. Camera A (Canon 7D) was outfitted with an 18-55 mm zoom lens at a 29 mm focal length and Camera B (Canon 7D) used an 18 mm fisheye lens. The GoPros (GP01, GP02, and GP03) were outfitted with the default fisheye lens. Recording was triggered with a hardwired camera switchfor Cameras A and B, while the GoPro Apple iOS application was used to wirelessly trigger the three GoPro cameras. A Nortec Vectrino Doppler current profiler was installed, centrally, within the flow tunnel with the sensor submerged in the water. Detailed information of the Small-Oscillatory Flow Tunnel can be found in Calantoni and others (2013).

Videos were captured by the five HD cameras. Video recording was started remotely, prior to the flow tunnel fly-wheel motor being turned on and prior to the Vectrino Profiler being started. On some occurrences, recording and/or the Vectrino were restarted, due to video length limitations or errors. For these instances, the fly-wheel motor was not stopped and recording began with the motor continuing to run.

Velocities were logged by a Vectrino acoustic Doppler profiler centrally-mounted within the tank. Velocity recording was remotely started from a connected computer after the beginning of the video recording. The start of the velocity record was indicated in the video by a single blink of a red LED indicator light, which was in the field of view of all cameras. 3D velocity was logged by the Vectrino software package and saved as raw binary files with file extension ".ntk" and a MATLAB version 2018b-readable format ".mat". The Vectrino acoustic velocity profiler measured the 3D water velocity (meters per second, m/s) of a 3.5 cm profile below the instrument, with 1-mm spaced bins and a sampling rate of 100 Hz.

Raw Vectrino data were processed in MATLAB by a script called "VectrinoII_ProcessData_2017.m" (modified NRL script), which extracted time, depth information, 3D velocity (m/s), and temperature from the ".mat" data files created by the Vectrino software. Noise and data spikes were removed from the 3D velocity data using two methods. First, the raw time-series data were compared to a smoothed time-series that was calculated by using an evenly-weighted convolution filter with a window size of 10. A standard deviation was calculated from the difference between the raw and smoothed time-series. Any raw sample with a difference greater than three standard deviations was replaced with the smoothed value. A new smoothed time-series was then calculated from the de-spiked time-series. This process was repeated two additional times using the de-spiked time-series in place of the raw data. Next, any sample recorded with a correlation (a measure of signal quality calculated by the Vectrino profiler software) less than 70% were replaced in the record with non-number (NaN) values. Quality controlled records of de-spiked and smoothed de-spiked velocity, temperature, distance from transducer, distance from bed, and sample time were saved in MATLAB readable ".mat" files.

Raw video frames were converted to individual individual Joint Photographic Experts Group (JPEG) image files utilizing MATLAB's built-in function, "VideoReader". Each frame of the raw video was read into MATLAB and saved as a unique, numbered and ordered JPEG image. The frames for Camera A and Camera B were rotated 180 degrees, since the cameras were mounted upside-down. The series of JPEG images were saved automatically to a unique directory named for the experiment date and run number identifier. The video frame rate was saved into a separate ".mat" file and saved into the same folder as the video image sequence.

Time differences between the video and corresponding velocity record were established by identifying the exact frame in which the Vectrino's red LED indicator light blinked in the camera's view. This instance was first estimated automatically by creating a unique digital mask around the LED for each camera. Since the cameras were fixed, the mask remained the same for every video. When a red flash was identified in the masked frame, that frame was saved to a ".mat" file unique to the experiment run. Once an estimated frame was found for all videos, the corresponding frames for each camera and each run were viewed to visually verify a red flash was captured. For instances were no flash was detected or an incorrect frame was picked, the correct frame was identified manually and the ".mat" file was updated. The time (in seconds) at which the Vectrino began recording in the video record was calculated as the frame number divided by the video frame rate. The start frame for the Vectrino in each video file is provided in 2017-329-FA_Data_Table.csv.

Flow velocity input for the video product was generated by extracting the along-tank component of flow velocity from the processed Vectrino 3D flow record. Free stream velocity was taken as an average of the Vectrino velocity bins, excluding any bins with erroneous velocities (such as constant zero or near zero values) or out of phase velocities compared to adjacent bins. These erroneous values might be a result of interference with the side of the tank, reflections off particles, or transfer issues. Transfer issues resulted when the computer used to record the Vectrino data in real-time was also performing additional tasks or data transfer. The bins used to calculate the average for each experiment run are listed in 2017-329-FA_Data_Table.csv. Due to storage limitations, excessively long Vectrino records were split into two files by the Vectrino software. These files were merged after processing into a continuous time-series.

Each flow velocity time-series was visually quality controlled (QC) for instances of transfer interference or tunnel maintenance which resulted in erroneous velocity values. Transfer errors occurred when data was transferred between disk drives on the desktop computer while the Vectrino was transferring data to the same computer. For the identified instances, the velocity was assigned a NaN value.

Interpretive video products were created in MATLAB v2018b and FFMPEG with an output file type of MPEG-4, using h264, at a FFMPEG constant quality of 15, a frame rate equal to the raw video frame rate, and resolution of 1080x1920 pixels. For each frame in the video, a figure was created consisting of the video frame from a single camera in the upper panel and the corresponding velocity time series in the bottom panel. A unique mask was created for each video camera to remove personnel and laboratory equipment not related to the transport of SOAs for privacy issues. Each mask was created using Matlab v2018b built in function "ginput" to create a polygon encompassing the small-oscillatory flow tunnel. Any pixel outside the polygon was assigned a red-green-blue pixel value of [0, 0, 0] (black). The video panel reduces the video resolution to approximately 830x1480. The velocity time series is displayed over a moving two-minute window with an orange dot indicating the current time in the video. The velocity time was adjusted to the video time by dividing the frame when the LED flash was detected in the video by the video frame rate. Only the velocity time series corresponding to collected video is shown. The current time indicator is centered in the bottom panel with the exception of the first minute and last minute of the video. The figure was saved as a portable network graphic ".png" file in a unique folder for each date, run, and camera. This was repeated for each frame in the video, then for each camera in a run, and then for all runs. A video was then created using FFMPEG for all frames for the unique date, run, camera combination using all frames in the selected folder. Each video has the format RunID_Camera.mp4, where RunID is a unique run identification code indicating a specific set of sediment, flow symmetry, and stroke length and is described in 2017-329-FA_Data_Table.csf. RunID ends with the Trial Number which consists of the letter, "T" followed by a sequential number starting at 1. Trial indicates the sediment type in the tank, flow symmetry, and stroke length remained unchanged. This occurred when the length of the video approached the limits of the camera's storage, there was concern regarding the Vectrino data quality, or a leak was observed in the tank. Camera is either CamA for Camera A, CamB for Camera B, GP01 for GoPro 1, or GP02 for GoPro 2. Trial indicates the sediment and tank configuration remained unchanged.

The video files were grouped by date and aSOA size prior to being archived within compressed ZIP (.zip) files. The filename convention utilized for each .zip file was: FAN_DATE_PT#, where FAN is the Field Activity Number (2016-364-DD), DATE is in the form YYYYMMDD where YYYY is the four-digit year, MM is the two-digit month, DD is the two-digit day, and PT# indicates the file grouping by sediment type for the day. Detailed information about which files are included in each archive, as well as sediment and tank settings, can be found in 2017-329-FA_Data_Table.csv.