The deep learning training of the GAIT dataset generated from the gait-logger-server and gait-logger-firmware are available in the src folder. The gait-deep learning mostly studies on using three architectures CNN, CNN-Statistics and CNN-LSTM whose implementation and training is done using a Keras API. Then using different parameters and benchmarking factors, the best model is generated, and converted using tensor-js library to be deployed to the Raspberry Pi for inference.
The overall deeplearning architecture for training all the data was trained using this particular flow. Email [email protected] for the full dataset.
The detailed implementation of CNN architecture is shown below:
The detail implementation of CNN-statistics is shown below:
the statistical features are computed using
And the final CNN architecture is given by
These were the results of the best models of
a. CNN
All the three architecture were deployed to the single board computer Raspberry Pi and the inference results were as follows: