Published September 17, 2024
| Version v1
Video/Audio
Open
Atmospheric Turbulence Dataset.
Authors/Creators
-
1.
University of Bristol
Description
This dataset comprises 38 video sequences that contain various levels of atmospheric turbulence.
Paul Hill, The University of Bristol
Please contact Paul Hill for any further information
Paul.Hill@bristol.ac.uk
Sequences
- The sequences were captured to give a range of
- Distances
- Turbulence levels
- Content (mainly aligning with discussed content of interest e.g. vehicles, buildings, roads etc.)
- The camera used was a Cannon R5, with a Canon 100-400mm zoom lens
- The sequences were captured over a period of three days in different locations
Preprocessing
- All the sequences were cropped in time to.
- Get manageable / easily processable sequences of length of approximately 20 seconds
- To remove unwanted artefacts (hands in front of lenses, people walking through the shot etc.)
- Cropped referred (in the Excel file) that they were spatially cropped from higher resolution sequences (to focus on more interesting / coherent content)
Sequence Descriptions
- The included Excel file: Heathaze_datasets_descriptions_2024.xlsx gives a detailed descriptiong each sequence.
Directory Structure
.
├── CLEAR1 # Static PNG image results of the CLEAR1 method
├── CLEAR2 # Example AVI CLEAR2 sliding window processed sequences
├── Original_Seqs # All Original Sequences in MP4 format
│ └── PNGs # The raw original sequences in directories named after the MP4 seq files.
- CLEAR1: Region and Pixel refer to the pixel by pixel and region-based fusion defined in the original CLEAR paper: Anantrasirichai et al. "Atmospheric Turbulence Mitigation Using Complex Wavelet-Based Fusion," Image Processing, IEEE Transactions on , vol.22, no.6, pp.2398-2408, June 2013
- CLEAR2: The outputs of an example CLEAR2 method output on the input sequences: Anantrasirichai et al. Atmospheric turbulence mitigation for sequences with moving objects using recursive image fusion", ICIP, pp. 2895-2899 2018.
- Original_Seqs: The PNGs are numbered in temporal order in each directory (named as per the MP4 seq files).