search

MikolajSzafraniecUPDS / shapedtw-python

Extension of dtw-python package implementing Shape DTW algorithm in two multidimensional variants
https://mikolajszafraniecupds.github.io/shapedtw-python/
GNU General Public License v3.0
16 stars 2 forks source link

Inquiry about Using shapeDTW for 2D Trajectory Matching #12

Closed arenas7307979 closed 9 months ago

arenas7307979 commented 10 months ago

Hi,

I hope this message finds you well. I am currently exploring the capabilities of the shapeDTW algorithm and have a specific use case in mind. My project involves working with 2D trajectories, and I am interested in understanding whether shapeDTW is suitable for matching these types of data.

Could you please clarify if shapeDTW can be effectively used for matching two-dimensional trajectories? If so, are there any particular considerations or modifications that I should be aware of to adapt the algorithm for 2D trajectory analysis?

Any insights or suggestions you could provide would be greatly appreciated.

Thank you for your time and assistance.

Best regards, arenas.

MikolajSzafraniecUPDS commented 10 months ago

Hi,

I trust this message finds you well. The matter at hand is non-trivial, and a definitive answer is challenging - I had to made a quick research. The authors of this paper (https://zheng-kai.com/paper/vldbj_2019.pdf) suggest it is possible to treating trajectories as two-dimensional time series. In such cases, shapeDTW can indeed be directly applied. However, it's crucial to note that treating trajectories as time series may not always be the optimal solution, making the problem more intricate.

The effectiveness of shapeDTW for 2D trajectory matching depends significantly on the nature of the input data. Dedicated preprocessing may be necessary based on the specific characteristics of your dataset.

If dealing with geographic data, as indicated in the article (https://alphasldiallo.github.io/dynamic-time-warping-algorithm-for-trajectories-similarity/), you might need to compute the distance matrix using the provided formula. This step ensures alignment with the requirements of the DTW algorithm. In such a case you can use 'dtw' package, which allows to use distance matrix directly as an input (currently there is no such possibility in 'shapedtw').

In conclusion, while shapeDTW can be a valuable tool for 2D trajectory analysis, careful consideration of the input data and potential preprocessing is essential. Feel free to reach out if you have further questions or require additional clarification.

Please let me know if this answer satisfies you.

Kind regards, Mikołaj