ludlows / csi-activity-recognition

Human Activity Recognition using Channel State Information

CSI-Activity-Recognition

Human Activity Recognition using Channel State Information for Wifi Applications

A simple Tensorflow 2.0+ model using Bidirectional LSTM stacked with one Attention Layer.

This code extends the previsous work of paper A Survey on Behaviour Recognition Using WiFi Channel State Information (corresponding code).

Dataset Preparation

Download the public dataset from here.

unzip the Dataset.tar.gz by the following command:

tar -xzvf Dataset.tar.gz

Inside the dataset, there are 7 different human activities: bed, fall, pickup, run, sitdown, standup and walk.

Requirements

Numpy

Tensorflow 2.0+

sklearn

Performance of the Model with Default Parameters

Default Parameters

Parameters for Batching Sequence	Value
window length	1000
Sliding Steps	200
Downsample Factor	2
Activity Present Threshold	0.6 (60%)

Parameters for Deep Learning Model	Value
# of units in Bidirectional LSTM	200
# of units in Attention Hidden State	400
Batch Size	128
Learning Rate	1e-4
Optimizer	Adam
# of Epochs	60

Model Architecture

Confusion Matrix

Label	Accuracy
bed	100%
fall	97.18%
pickup	98.68%
run	100%
sitdown	95%
standup	95.56%
walk	99.51%

Usage

Download the code from github.

git clone https://github.com/ludlows/CSI-Activity-Recognition.git 

Enter the code folder.

cd CSI-Activity-Recognition

Run The Model with Default Parameters

python csimodel.py your_raw_Dataset_folder

Meanwhile, you could also modify the parameters in the csimodel.py or change the architectures of neural networks.

This code could be a starting point for your deep learning project using Channel State Information.