Original dataset available at: https://icu.ee.ethz.ch/research/datsets.html
[For now this tool does not support the Hotel dataset sequence yet.]
Step 0: Install dependencies
python3 -m pip install opencv-python numpy matplotlib tqdm
Step 1: Preprocessing
cd ewap_tools/seq_eth/
mkdir frames
mkdir images
ffmpeg -i seq_eth.avi -vf fps=25 frames/frame%08d.png
This process may take several minutes to finish, when it is done, go to the directory "ewap_tools/seq_eth/frames/", you should be able to see every frame of the original video as a png file.
Step 2: Run the example
python3 example.py
Feel free to check out the documented source code in reader.py to get more information.
The orignal format of the dataset can be found at ewap_tools/ewap_dataset/README.txt
The reader code preprocesses the orignal data at ewap_tools/ewap_dataset/seq_eth/obsmat.txt, and store the state of a pedestrian at a frame as a 4-vector: [position_x, position_y, velocity_x, velocity_y]
The time step between two frames in the original dataset is 0.4s, we interpolate the original data and the time step between two frames in the reader code is 0.1s.
Muchen Sun (muchen@u.northwestern.edu)