Skip to content

Latest commit

 

History

History
99 lines (63 loc) · 6.77 KB

README.md

File metadata and controls

99 lines (63 loc) · 6.77 KB

Accelerometer sensor dataset for Human Posture Recognition

License: GPL v3

https://github.com/pkmandke/Human-Posture-Dataset/blob/master/img/sample_pose.png

This repository links to the dataset that we created during our work on Human Posture Recognition at College of Engineering, Pune in the Spring of 2018. The dataset consists of accelerometer sensor values for 6 postures namely - standing, sitting, sleeping, running, forward bending and backward bending - from the Invensense' MEMS IMU sensor MPU-6050 mounted one each on the left chest and the right thigh.

The dataset is in the form of a single csv file of size 3.2MB. Download.

Data Summary

Posture Number of samples
Sleeping 8329
Standing 8358
Sitting 7542
Running 3617
Forward Bending 11504
Backward Bending 5450
Total 44800

Details

https://github.com/pkmandke/Human-Posture-Dataset/blob/master/img/b_diag.png

Data

The csv file has a total of 44800 samples each having 6 columns. The Ax1, Ay1 and Az1 are the accelerometer readings from the chest sensor while Ax2, Ay2 and Az2 are the readings from the thigh sensor. The readings are not calibrated or filtered and contain any noise as is. The last column of labels identifies the corresponding posture. The posture legend is as follows:

Label Posture
0 Sleeping
1 Standing
2 Sitting
3 Running
4 Forward Bending
5 Backward Bending

Nuances

The data has been collected from a total of 3 subjects who were students at College of Engineering, Pune. Note that the readings for a given posture from all 3 subjects have been combined and are indistinguishable. Please contact us if you need the readings of each subject and posture separately.

We encourage you to shuffle the data for a single posture for robustness during training a classifier. Moreover, the number of readings for each posture and subject are not the same. This is because of 2 reasons. Firstly, we removed outliers from each data collection session. Secondly, it was difficult to maintain the same postures for long periods of time and we ensured the comfort of the subjects.

https://github.com/pkmandke/Human-Posture-Dataset/blob/master/img/sensor_node.png

In order to collect the data, we built a small PCB to interface the MPU-6050 with the Espressif's ESP8266 based NodeMCU and deployed the sensor on the human subjects as shown above. The data was then wirelessly transmitted from the NodeMCU to a central node interfaced serially with a PC where the readings were sampled at 200ms intervals. HTTP was used over TCP/IP for the data transfer to and from the NodeMCU.

During the static postures, the subjects were directed to include slight natural movements which make the readings more robust for training classifiers. For the bending postures, the subjects were in sitting position and bent forward and backward at 30-40 degrees to the vertical. For running, the subjects jogged in the same place for the most part with slight translational motion or displacement occasionally. Thus, the readings for running are spurious at best.

For more details, refer to our paper here. Feel free to contact us for further queries.

Data visualization

https://github.com/pkmandke/Human-Posture-Dataset/blob/master/img/sleephist.png

Scripts in the form of ipython notebooks that vizualize the data and also train sample classifiers will be released soon. Stay tuned!

Citation

If you use our dataset in your work, please cite our paper.

H. Kale, P. Mandke, H. Mahajan and V. Deshpande, "Human Posture Recognition using Artificial Neural Networks," 2018 IEEE 8th International Advance Computing Conference (IACC), Greater Noida, India, 2018, pp. 272-278.

@INPROCEEDINGS{8692143, author={H. {Kale} and P. {Mandke} and H. {Mahajan} and V. {Deshpande}}, booktitle={2018 IEEE 8th International Advance Computing Conference (IACC)}, title={Human Posture Recognition using Artificial Neural Networks}, year={2018}, volume={}, number={}, pages={272-278}, keywords={health care;image recognition;neural nets;wireless LAN;human posture recognition;human postures;healthcare analysis;optimal neural network architecture;artificial neural networks;invasive approach;intrusive approach;Raspberry-Pi device;Wi-Fi;pattern-recognition;neural-networks;posture-recognition;wearable-sensors;accelerometers;embedded systems;real time movement classification;sensor positioning}, doi={10.1109/IADCC.2018.8692143}, ISSN={2164-8263}, month={Dec},}

Authors

License

TL;DR - Feel free to use this in your work so long as you cite us!

This dataset is licensed under the GPL license. See the license for details.

Acknowledgements

We express our sincere gratitude towards our guide Mrs. Deeplaxmi Niture, Assistant Professor at the Department of Electronics and Telecommunication, College of Engineering, Pune (COEP) for her help in publishing our work. We also thank the Department of Electronics and Telecommunocation at COEP as well as the COEP Amateur Radio Club for providing the resources and lab space for our work.