"Quantifying the degradation of optical algorithms in increasingly turbid mediums."
Datasets used for: Mitchell Scott and Aaron Marburg. Quantifying the degradation of optical algorithms in increasingly turbid mediums. In OCEANS 2021 San Diego. IEEE, 2021. These datasets were collected in the UW-APL test tank using either a Numurus 3DX-C stereo camera unit or a custom Dalsa stereo camera pair. Data were recorded using ROS noetic.
Calibration information from this optical dataset is included in the below links as .yaml files.
Dataset #1
This dataset contains rosbags of optical images and depth maps produced by our Numurus 3DX-C unit without motion. Turbidity was induced over 13 steps using cornstarch. An info.txt is present in each directory indicating the turbidity level at each step.
The dataset is posted here.
Figure 1: Photo of Electrical Flying Lead (EFL) in UW-APL tank in no turbidity. Taken with Numurus unit.
Figure 2: Photo of Numurus-generated depth map of EFL in UW-APL tank with no turbidity.
Dataset #2
This dataset contains .pngs of optical images over four different sweeps ranging from no turbidity to high turbidity. Associated labelme annotations of select images and their corresponding Darknet style .txt files are included for deep learning training.
The dataset is posted here.
Figure 3: Photo of EFL in UW-APL tank in low turbidity. Taken with custom Dalsa machine vision camera stereo pair.