gaitutils library

Useful code to generate and manipulate data for gait recognition.

In particular, this library offers the following functionality:

Person detection and tracking in videos.
Optical flow computation.
Generation of person-centric optical flow samples. Ready to be used with OF-based CNNs for gait recognition.

Demo code at Google Colab

1) People detection and tracking

Use the following Google Colab:

2) Optical flow computation

Use the following Google Colab:

3) Generate OF-based samples for gait recognition

This step has to be performed after 1 and 2.

You can use the latest section of the previous Colab (in step 2) to generate the actual input samples for the CNN.

References

If you either use this code or find useful this repository, please, cite any of the following related works:

[A] Francisco M. Castro, Manuel J. Marín-Jiménez, Nicolás Guil, Santiago Lopez Tapia, Nicolas Pérez de la Blanca:
    Evaluation of CNN Architectures for Gait Recognition Based on Optical Flow Maps. BIOSIG 2017: 251-258
[B] Rubén Delgado-Escaño, Francisco M. Castro, Julián Ramos Cózar, Manuel J. Marín-Jiménez, Nicolás Guil:
    MuPeG - The Multiple Person Gait Framework. Sensors 20(5): 1358 (2020)
[C] Francisco M. Castro, Manuel J. Marín-Jiménez, Nicolás Guil, Nicolás Pérez de la Blanca:
    Multimodal feature fusion for CNN-based gait recognition: an empirical comparison. Neural Comput. Appl. 32(17): 14173-14193 (2020)
[D] R. Delgado-Escaño, F. Castro, N. Guil, V. Kalogeiton, M. Marín-Jiménez:  
    Multimodal gait recognition under missing modalities. IEEE ICIP, 2021