Seq2Seq RNN based Gait Anomaly Detection from Smartphone Acquired Multimodal Motion Data

Deep learning approach to anomaly detection in gait data acquired through smartphone's multimodal sensors. The proposed architecture takes advantage of RNN and CNN layers to implement a Sequence-to-Sequence feature extractor and a Convolutional classifier, check the paper for more details.

Welcome

If you find any piece of code valuable for your research please cite this work:

And don't forget to give us a ⭐ in the GitHub banner 😉.

Project structure

The folders are organized as follows:

• ALV contains the Android application ActivityLoggerVideo used for the data collection.
• pre-processing contains the scripts to apply the preprocessing transformations of the signals described in the paper. It is the necessary data preparation step for the later use in the deep learning framework.
• Seq2Seq-gait-analysis contains the inference code and details on how to use it.
• Seq2Seq-gait-analysis/Data is a placeholder for the preprocessed gait sequences used for training and testing of the proposed network.
  

Check each folder's README for more details.

License

The code is released under the MIT License.