How linearly do people walk?
This repo. analyses the MOT-17 and MOT-20 datasets to find out if people's motions are piecewise linear, and implements some simple methods for detecting piecewise-linear motion, as well as some random walk baselines.
Linear segments identified from the piecewise linear analysis. Blue points indicate the points in the original trajectory. Red lines indicate the linear segments. The tolerance used is 0.1 for the maximum allowed different between neighboring slopes for the point in the center to be considered as part of a linear segment.
Read about this project on Medium.
Install the included motlinearity helper package:
pip install -r requirements.txt
pip install -e .Download the MOT-17 data https://motchallenge.net/data/MOT17/ (and possibly MOT20). The data should be located in analysis/MOT17Labels/....
cd analysis
python analyze.py --mot MOT17 --command <XXX>You can use python analyze.py --help for a more complete list of options, summarized here:
(1) plot-traj - Plot some trajectories from the dataset and their linear segments.
(2) plot-traj-tog - Plot some trajectories from the dataset and their linear segments side-by-side.
(3) lin-analysis - Run the linear analysis for the dataset.
(4) random-walk-sim - simulate a random walk.
(5) random-walk-analysis - Analyze the random walk.
(6) plot-traj-tog-random-walk - Plot the trajectories and linear segments from the random walks.