This software annotates .c3d gait trajectories with Heel-Strike and Foot-Off events. It uses neural networks (more precisely Long Short Term Memory networks) through keras and tensorflow packages.
This repository contains the training code. If you want to use pretrained version for your data refer to https://github.com/kidzik/event-detector
This repository contains three key procedures:
- Convert
.c3dto.csvwith frames as rows and markers/joint angles as columns - Train with different hyperparameters
- Compare the best models with existing marker-based models
These procedures are implemented in the following files respectively:
- load.py
- experiments.py
- Comparison.ipynb
Python notebook for training models interactively (with extended comments): https://github.com/kidzik/event-detector-train/blob/master/Training.ipynb
Implementation of heursitic based methods: https://github.com/kidzik/event-detector-train/blob/master/heuristics.py
This research was sponsored by the Mobilize Center, a National Institutes of Health Big Data to Knowledge (BD2K) Center of Excellence supported through Grant U54EB020405. The model is trained on the data from Gillette Children's Specialty Healthcare, in accordance with the data sharing agreement.