yupidevs / yupi

Python package designed for collecting and processing trajectory data.

Standing for Yet Underused Path Instruments, yupi is a set of tools designed for collecting, generating and processing trajectory data.

Main features

• Convert raw data to trajectories ... different input manners
• I/O operations with trajectories ... json and csv serializers
• Trajectory extraction from video inputs ... even with moving camera
• Artificial trajectory generation ... several models implemented
• Trajectory basic operations ... rotation, shift, scaling, ...
• Trajectory transformations ... filters, resamplers, ...
• Statistical analysis rom trajectories ensembles ... turning angles histogram, velocity autocorrelation function, power spectral density, and much more ...
• Results visualization ... each statistical observable has a related plot function
• Spacial projection of trajectories ... for 2D and 3D trajectories

Installation

Current recommended installation method is via PyPI:

pip install yupi

Compatibility

• Python 3.7 or later
• Ubuntu 18.04 or later
• Windows 7 or later
• macOS 10.12.6 (Sierra) or later.

Getting Started

In the official documentation there are some resources to start using the library: Tutorials, Examples and a detailed description of the API.

Examples

Code examples (with additional multimedia resources) can be found in this repository. Additionally, in the Examples section of the documentation, you can find the same examples with additional comments and expected execution results in order to inspect the examples without actually executing them.

yupi's Issues

Error when loading previously saved trajectories without uniformly sampled time data

Hi and thanks for your work! I have a question regarding trajectories which are not uniformly sampled in time. Reading your "brief comment on time" in the documentation I thought I only have to supply a sequence of time data and optionally set dt to None. However I realized the dt value set is on creation of the Trajectory object.

points1 = [[0, 0], [1.0, 0], [0.63, 0.98]]
t1 = [0,0.5,2]
track1 = Trajectory(points=points1, t=t1,traj_id="track1")
print(track1.dt)

points2 = [[0, 0], [2.0, 0], [1, 0.98]]
t2 = [0.5,1,1.3]
track2 = Trajectory(points=points2, dt=None, t=t2,traj_id="track2")
print(track2.dt)

Trajectory.save_trajectories([track1,track2], "./demo")
trajs = Trajectory.load_folder("./demo")

output:

1.0
0.4
Ignoring: demo/trajectory_1.json
Ignoring: demo/trajectory_0.json

When saving the trajectories and loading them again this leads to a LoadTrajectoryError triggered by a ValueError:
ValueError: You are giving 'dt' and 't' but 't0' is not the same as the first value of 't'.

Am I understanding something wrong or is this a bug in the implementation?

Add `plot_trajectory` function to `__init__.py` on analyzing module

Currently the __all__ variable on the __init__.py file of the analyzing module only contains the plot_trajectories function.

Implement add offset of a trajectory

Add a constant offset in polar coordinates for trajectories of 2 dims

Check analyzing functions for non uniformly spaced trajectories

Add examples

Great workl! Please add some working examples

Document new Trajectory creation methods

Since, now we have 3 different ways to create a Trajectory object from data, we should include a demo of each in:

• Getting started section of the docs:
  Subsection: 'Converting your data into Trajectory objects'
• Intro001.py of the examples should be extended to contain all 3 ways.

Link API resources in examples.rst to their respective page in API reference

Now there are some resources that are not being linked automatically. Maybe, we need to generate the links elsewhere.

Add `recursively` parameter in Trajectory's `load_folder` method

Right now this method loads all the trajectories from a folder searching in a recursively way. Lets add a parameter so the user can control this behavior.

Feature Request: Linking Tracklets to Trajectories

Not sure how much features you want to integrate into yupi but I thought it would be really great to have general tracklet linking features in order to connect tracklets created due to id switches while tracking.

Thanks for all your work on yupi!

Change velocity calculation on trajectory

When time data is given the velocity must be calculated using this data. At this moment is calculated using the mean of the differences (average dt).

Unable to overwrite time or component attributes of Trajectory objects

After creating a Trajectory object, traj, it is possible to successfully overwrite traj.r by using +=. For instance:

import yupi
x = [2, 5, 3]
traj = yupi.Trajectory(x)

traj.r += 1
traj.r
# Vector([[3.],
          [6.],
          [4.])

However, one gets an Attribute Error when trying to do the analogous change with time or component attributes, as indicted bellow:

traj.t += 1
# Attribute Error: can't set attribute 't'

traj.r.x += 1
# Attribute Error: can't set attribute 'x'

It would be great if one could overwrite other Trajectory's attributes as well.

Implement trajectory slicing

Being able to do:

t2 = t1[50:200]

Add error handling when loading trajectories

Add new tracking methods

New tracking methods to be implemented:

• Frame Diff
• Template Matching
• Background subtraction
• Feature detection

Add vector hash

Vector hashes will be used to check if some trajectory data must be recalculated.

Add type annotations

Error creating an undistorted

The bug can be reproduced by executing example_004.py from here:

https://github.com/yupidevs/yupi_examples/blob/master/example_004.py

Implement + and - of trajectories