Since 1870 a large growth in human life expectancy has been observed in Europe. This growth has expanded in the whole world principally due to the great achievements in health care field. As a result, the proportion of elderly people is rapidly increasing. Aging people in general lives in isolated conditions. In addition to that some of them are not capable to live normally and take advantages from health care facilities services. Building remote monitoring systems for elderly patients who live alone or without permanent caretaking will improve their quality of life. For better decision making these remote monitoring systems needs some regular and trustful information about patients.
To fulfil remote monitoring systems’ requirements Jorge Luis Reyes Ortiz has developed a complete Human Activity Recognition System able to detect and recognize 12 different activities performed by humans in their daily living in online mode using smartphones. The recognition part of his system is based on an SVM model already trained, capable of predicting activities performed by users. Necessary datasets of users’ movements will be collected from smartphone sensors (accelerometer and gyroscope), processed and then fed to the prediction model to recognize performed activities.
To build the final model embedded in this HAR system. An offline version of this system needs to be created for two reasons:
2. To build the train-test pipelines and test different state of art ML alogrithms to choose the optimal one.
The goal of this project is to build a machine learning model and a signal processing pipeline in offline mode capable of processing signals collected using smart phone inertial sensors and producing useful datasets will be used as inputs of a machine learning model capable of recognizing some of human daily activities (sitting, walking …) included in the dataset (see datasets and Inputs section) with a low error rate. The signal processing pipeline and the final model could be used as a good source of information about patient’s daily activities needed by remote monitoring systems mentioned earlier
This project requires Python 3.5 and the following Python libraries installed:
- Python 3.5
- NumPy , SciPy , Pandas , matplotlib
- Statsmodel , Spectrum , scikit-learn
We recommend to install Anaconda, a pre-packaged Python distribution that contains some of the necessary libraries and software for this project (Be sure to install an Anaconda environment that includes python 3.x). You will also need to have software installed to run and execute an iPython Notebook
Here is an YAML File that includes environement dependencies used in this project requirements you shoud use it to generate the anaconda enviroment.
This repository Human-Activity-Recognition-Using-Smartphones includes three main directories and 3 files:
- `\Description-Images\`: This folder includes images used for aesthetic purposes.
- `\Reports\`: Contains an HTML version of each notebook + 2 reports
- `.\Data\`: Under this Directory you will find all Datasets used in this project.
- `.\Original-Data\`: contains the original Datasets brought by Jorge Luis Reyes Ortiz.
- `. \UCI-HAR-Dataset\`: The first version of this Dataset V 1.0.
- `. \HAPT-Dataset\`: The second version V 2.1.
-`.\ New Data\`: Under this directory you will find the processed datasets generated from the original
ones using: `Part_I--Signal-Processing-Pipeline.ipynb`.
- `Part_I--Signal-Processing-Pipeline.ipynb`: This notebook file contains the signal processing
pipeline, visualizations and detailed analysis of each step
performed.
- `Part_II--Machine-Learning-Part.ipynb`: This is the notebook file which related to machine learning,
visualizations and detailed analysis of each step performed
to build the final model.
- `README.md` : It contains a short description of this project and necessary steps to
run it successfully.
The experiments were carried out with a group of 30 volunteers within an age bracket of 19-48 years. They performed a protocol of activities composed of six basic activities: three static postures (standing, sitting, lying) and three dynamic activities (walking, walking downstairs and walking upstairs). The experiment also included postural transitions that occurred between the static postures. These are: stand-to-sit, sit-to-stand, sit-to-lie, lie-to-sit, stand-to-lie, and lie-to-stand.
All the participants were wearing a smartphone (Samsung Galaxy S II) on the waist during the experiment execution. They captured 3-axial linear acceleration and 3-axial angular velocity at a constant rate of 50Hz using the embedded accelerometer and gyroscope of the device. The experiments were video-recorded to label the data manually. The resulted dataset stored in the directory Raw-Data could be considered as the original labelled dataset and from it different subsets would be generated.
| User 01 inertial signals |
|---|
 |
| Walking | Standing |
|---|---|
 |
 |
| Sitting | Lying |
|---|---|
 |
 |
This Raw-Data was randomly partitioned into two sets, where 70% of the volunteers was selected for generating the training data and 30% for the test data.
Noise filtering:
The features selected for this database come from the accelerometer and gyroscope 3-axial raw signals t_Acc-XYZ and t_Gyro-XYZ. These time domain signals (prefix 't' to denote time) were captured at a constant rate of 50 Hz. Then they were filtered using a median filter and a 3rd order low pass Butterworth filter with a corner frequency of 20 Hz to remove noise.
1- Median Filter: was applied to reduce background noise.
2- 3rd order Low pass Butterworth filter with a cut-off, frequency = 20hz was applied to remove high frequency noise.
Resulted Signals are: total_acc_XYZ and Gyro_XYZ.
Similarly, the acceleration signal total-acc-XYZ was then separated into body and gravity acceleration signals (tBodyAcc-XYZ and tGravityAcc-XYZ) using another low pass Butterworth filter with a corner frequency of 0.3 Hz. Since the gravitational force is assumed to have only low frequency components.
Resulted components are: total_acc-XYZ ==> tBody_acc-XYZ + tGravity_acc-XYZ
From the processed signals presented in the picture above, they extract data points and activity labels related only to the first six activities which are: standing, sitting, lying, walking, walking downstairs and walking upstairs called Basic Activites(BA) to build the first version of this dataset.
**The second version of this datasets includes all the 12 activities : Basic Activities (BA) and Postural Transitions (PT).
Subsequently, the body linear acceleration and angular velocity were derived in time to obtain Jerk signals (tBodyAccJerk-XYZ and tBodyGyroJerk-XYZ). Also, the magnitude of these three-dimensional signals was calculated using the Euclidean norm (tBodyAccMag, tGravityAccMag, tBodyAccJerkMag, tBodyGyroMag, tBodyGyroJerkMag).
Finally, a Fast Fourier Transform (FFT) was applied to some of these signals producing fBodyAcc-XYZ, fBodyAccJerk-XYZ, fBodyGyro-XYZ, fBodyAccJerkMag, fBodyGyroMag, fBodyGyroJerkMag.
The f to indicate frequency domain signals.
These Signals were then sampled in fixed-width sliding windows of 2.56 sec and 50% overlap (128 readings/window).
From each sampled window, a vector of 561 features was obtained by calculating variables from the time and frequency domain.
Jorge L. Reyes-Ortiz(1,2), Davide Anguita(1), Alessandro Ghio(1), Luca Oneto(1) and Xavier Parra(2):
[1] - Smartlab - Non-Linear Complex Systems Laboratory DITEN - Università degli Studi di Genova, Genoa (I-16145), Italy.
[2] - CETpD - Technical Research Centre for Dependency Care and Autonomous Living Universitat Politècnica de Catalunya (BarcelonaTech). Vilanova i la Geltrú (08800), Spain.
@mail: [email protected]
This first version is located in ./Data/Original-Data/UCI-HAR-Dataset/. As mentioned earlier this dataset concerns only Basic Activities performed by the users during the experiments. It includes two main directories and they can be used separately.
Dataset Architecture:
Under The UCI-HAR-Dataset Directory we have:
-
./Inertial Signals/: This directory includes the Semi-processed features of this version.- ` ./Inertial-Signals/train/`: The train folder includes 11 files. - `total_acc_x_train.txt`: The acceleration signal from the smartphone accelerometer X axis in standard gravity unit 'g'. Every row shows a 128-element vector. The same description applies for the `total_acc_y_train.txt` and `total_acc_z_train.txt` files for the Y and Z axis. - `body_acc_x_train.txt`: The body acceleration signal obtained by subtracting the gravity from the total acceleration. The same description applies for the `body_acc_y_train.txt` and `body_acc_z_train.txt` files for the Y and Z axis. - `body_gyro_acc_x_train.txt`: The angular velocity vector measured by the gyroscope for each window sample. The units are radians/second. The same description applies for the `body_gyro_y_train.txt` and `body_gyro_z_train.txt` files for the Y and Z axis. - ` ./Inertial-Signals/test/*`: This folder includes necessary testing files of inertial signals following the same analogy as in `./Inertial Signals/train/`. -
./Processed-Data/: This directory includes the fully processed features which concerns the same six activities.- `X_train.txt`: Train features, each line is composed 561-feature vector with time and frequency domain variables. - `X_test.txt`: Test features, each line is composed 561-feature vector with time and frequency domain variables. - `features_info.txt`: Shows information about the variables used on the feature vector. - `features.txt`: includes list of all 561 features
-
y_train.txt: train activity labels, Its range is from 1 to 6 -
y_test.txt: test activity labels, Its range is from 1 to 6 -
subject_train.txt: training subject identifiers, Its range is from 1 to 30 -
subject_test.txt: testing subject identifiers, Its range is from 1 to 30 -
activity_labels.txt: -
README.md:
Licence:
Use of this dataset in publications must be acknowledged by referencing the following publication [1]:
[1]- Davide Anguita, Alessandro Ghio, Luca Oneto, Xavier Parra and Jorge L. Reyes-Ortiz. A Public Domain Dataset for Human Activity Recognition Using Smartphones. 21th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, ESANN 2013. Bruges, Belgium 24-26 April 2013.
Note: This dataset is distributed AS-IS and no responsibility implied or explicit can be addressed to the authors or their institutions for its use or misuse. Any commercial use is prohibited.
Jorge L. Reyes-Ortiz(1,2), Davide Anguita(1), Luca Oneto(1) and Xavier Parra(2)
[1] - Smartlab, DIBRIS - Università degli Studi di Genova, Genoa (16145), Italy.
[2] - CETpD - Universitat Politècnica de Catalunya. Vilanova i la Geltrú (08800), Spain.
@mail: [email protected]
Website: www.smartlab.ws
This dataset is an extended version of the UCI Human Activity Recognition Using smartphones Dataset V 1.0 mentioned earlier.
This version provides the original signals captured from the smartphone sensors mentioned earlier Raw-Data, instead of the ones semi-processed and sampled into windows located in Inertial-signals directory which were provided in version 1.0. This change was done in order to be able to make online tests with the data. Moreover, the activity labels were updated in order to include postural transitions that were not part of the previous version of the dataset. This Dataset includes two main directories and they can be used separately.
Dataset Architecture:
Under The HAPT-Dataset Directory we have:
-
./Raw-Data/: This the original data (unprocessed data) obtained directly from the set of experiments after activity labelling . It includes Raw triaxial signals from the accelerometer and gyroscope of all trials with participants. This Directory includes 123 files:- **61 acc files are**: `Raw-Data/acc_expXX_userYY.txt`: The raw triaxial acceleration signal for the experiment number XX and associated to the user number YY. Every row is one acceleration sample (three axis)[X,Y,Z] captured at a frequency of 50Hz. - **61 gyro files are**: `Raw-Data/gyro_expXX_userYY.txt`: The raw triaxial angular speed signal for the experiment number XX and associated to the user number YY. Every row is one angular velocity sample (three axis)[X,Y,Z] captured at a frequency of 50Hz. - **One last file is** :`RawData/labels.txt`: includes labels of all the performed activities (1 per row). - Column 1: experiment number ID, - Column 2: user number ID, - Column 3: activity number ID - Column 4: Label start point (in number of signal log samples (recorded at 50Hz)) - Column 5: Label end point (in number of signal log samples)
| Raw Data Samples |
|---|
 |
| Number of rows per Experience | Number of rows per activity |
|---|---|
 |
 |
| Number of useful rows per Experience | Mean time in seconds per Activity |
|---|---|
 |
 |
-
./ Processed-Data/: This Directory includes the fully processed dataset which concerns all The twelve activities performed by users in the experiment.- `X_train.txt` : Train features, each line is composed 561-feature vector with time and frequency domain variables. - `X_test.txt` : Test features, each line is composed 561-feature vector with time and frequency domain variables. - `y_train.txt`: train activity labels, its range is from 1 to 12 - `y_test.txt`: test activity labels, its range is from 1 to 12 - `subject_id_train.txt`: training subject identifiers, Its range is from 1 to 30 - `subject_id_test.txt`: testing subject identifiers, Its range is from 1 to 30 - `features_info.txt`: Shows information about the variables used on the feature vector. - `features.txt`: includes list of all 561 features
activity_labels.txt:README.md:
License:
Use of this dataset in publications must be acknowledged by referencing the following publications [1]:
[1] - Jorge-L. Reyes-Ortiz, Luca Oneto, Albert Samà, Xavier Parra, Davide Anguita. Transition-Aware Human Activity Recognition Using Smartphones. Neurocomputing. Springer 2015.
Note: This dataset is distributed AS-IS and no responsibility implied or explicit can be addressed to the authors or their institutions for its use or misuse. Any commercial use is prohibited.
The New Datasets are the results of applying my own signal processing pipeline to Raw-Data. This directory includes one main folder full_Datasets_type_I_and_II:
Type I : to denote V 1.0 which includes Basic activities only.
Type II : to denote V 2.1 which includes Both Basic Activities and Postural Transitions.
-
./New-Data/full_Datasets_type_I_and_II/: includes Datasets produced by the signal processing pipeline usingRaw-Dataonly. All parts are fully processed. Each line(observation) includes XXX features + the subject_Id and the activity label related the observation.- ` Dataset_I_part1.csv `: The first 5000 rows of Dataset type I - ` Dataset_I_part2.csv `: The rest of rows of Dataset type I - ` Dataset_II_part1.csv `: The first 6000 rows of Dataset type II - ` Dataset_II_part2.csv `: The rest of rows of Dataset type II-
new features info.txt: Includes info about features(columns) in both Datasets type I and type II -
new_features.txt: the full list of features (column names) of Dataset type I and II
-
In the Terminal or Command Prompt, navigate to the folder containing the project files which contains this README.md and then use the command jupyter notebook file_name.ipynb to open up a browser window or tab to work with your notebook. Alternatively, you can use the command jupyter notebook or ipython notebook and navigate to the notebook file in the browser window that opens.
If you want to generate RawData statistics and overwrite the new datasets type I and II you should run the Part_I--Signal-Processing-Pipeline.ipynb.
If you want to generate new datasets type I and II statistics and its machine learning results you should run the second notebook Part_II--Machine-Learning-Part.ipynb.
Notes:
-
Please before running any notebook pay attention to running durations first (durations were computed on a laptop core i5 8Gbs of RAM).
-
The Data owners didn't provide the signal processing code. The signal processing pipeline was developed using only explanations provided in both original datasets (Version 1.O and 2.1). As a result The General Process mentioned earlier is a little bit different from the one coded in the signal processing pipeline.
-
The windowing methods used in the signal processing pipeline differs depending on the type of the data generated. Reports included in this reporsitery provides detailed information about these differences.
Please If you have any ambiguity about any part of this project or just want to have more info about this subject don't hestitate to contact me at: [email protected]