TRACE

TRACE-IMU (Trajectory Classification Employing IMU) is a combination of machine learning techniques in order to first generate poses in SMPL format from six IMU sensors using DIP, which are then tracked over time to form trajectories. These trajectories can be classified either as time series or images.

Usage

For own data collection e.g. with GaitUp Physilog sensors, check the jupyter notebook Preprocessing_Physilog.ipynb.

DIP

1. Clone https://github.com/eth-ait/dip18 and cd into train_and_eval
2. Download pretrained model and DIP-IMU dataset from http://dip.is.tuebingen.mpg.de/downloads into models and data directories
3. Install dependencies

conda create --name name_of_your_choice_dip python=3.5
conda activate name_of_your_choice_dip
pip install --ignore-installed --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.4.0-cp35-cp35m-linux_x86_64.whl
conda install opencv
pip install numpy-quaternion
conda install numba

4. Generate poses from IMU dataset

python run_evaluation.py --system local --data_file own --model_id 1528208085 --save_dir ./models --eval_dir ./evaluation_results --datasets dip-imu --past_frames 20 --future_frames 5 --save_predictions

Conversion to TRACE-compatible dataset

1. Download SMPL body models from https://mano.is.tue.mpg.de/downloads
2. Generate Raw Sensor Dataset

python generate_acc_ori.py dip_imu

3. Generate Trajectory Dataset

python generate_trajectories.py dip_imu
python transform_trajectories.py dip_imu

TRACE

Dependencies

conda create --name name_of_your_choice_trace python=3.7
conda activate name_of_your_choice_trace
conda install -c anaconda pandas scikit-learn tensorflow-gpu=2.1.0
conda install -c conda-forge keras opencv
pip install git+https://github.com/nghorbani/configer
pip install git+https://github.com/nghorbani/human_body_prior

change line 506 of /path/to/env/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py from _LOCAL_DEVICES = tf.config.experimental_list_devices() to _LOCAL_DEVICES = [x.name for x in tf.config.list_logical_devices()]

Start Training

The entrypoint for training is train.py. The parameters are explained executing python train.py --help. An example call looks like this:

python train.py --dataset dip_imu --model ConvLSTM --data_path /path/to/data/ --gpu 1 --horizontal_flip

Classification