Sonar for your Home Assistant

The report can be found here.

The presentation slides can be found here.

This project was done by Seraphine Goh and Dennis Shim in Spring 2019 for EE209AS: Security and Privacy for Embedded Systems, Cyber-Physical Systems, and Internet-of-Things at UCLA.

Data

Data can be found at this link.

Below is a table describing the data.

Data Directory	Frequency of sound	Description
`trials_no_subject_1`	1 kHz	No subject present, 50 trials in 1 furniture configuration
`trials_no_subject_2`	1 kHz	No subject present, 100 trials in 4 furniture configurations
`trials_no_subject_3`	1 kHz	No subject present, 105 trials in 47 furniture configurations
`trials_no_subject_4`	1 kHz	No subject present, 100 trials in 1 furniture configuration
`trials_no_subject_5`	1 kHz	No subject present, 100 trials in 1 furniture configuration
`trials_front_1`	1 kHz	Subject in front, 100 trials sitting and standing in 1 furniture configuration
`trials_front_2`	1 kHz	Subject in front, 100 trials sitting in multiple configurations in 1 furniture configuration
`trials_front_3`	1 kHz	Subject in front, 200 trials in front, sitting on couch (1-60) and standing (61-200) (moving around as well). 4 different furniture configurations
`trials_behind`	1 kHz	Subject behind, 100 trials of standing, 2 different furniture configurations (50 each)
`trials_right`	1 kHz	Subject right, 200 trials of sitting (15 * (2 furniture configurations) * (3 seating arrangements) and standing (5 furniture configurations)
`trials_left`	1 kHz	Subject left, 200 trials of sitting (15 * (2 furniture configurations) * (3 seating arrangements) and standing (5 furniture configurations)
`20kHz_trials_no_subject`	20 kHz	No subject, 210 trials of 14 furniture configurations
`20kHz_trials_front`	20 kHz	Subject in front, 200 trials of sitting/standing and multiple furniture configurations
`100Hz_trials_no_subject`	100 Hz	No subject, 200 trials of >10 furniture configurations
`100Hz_trials_front`	100 Hz	Subject in front, 200 trials of sitting/standing and multiple furniture configurations
`trials_moving_ccw`	1 kHz	Subject walking around microphone/speaker setup counterclockwise at different speeds, different paths, and different start/stop points
`trials_moving_cw`	1 kHz	Subject walking around microphone/speaker setup clockwise at different speeds, different paths, and different start/stop points

Usage

Data Recording

Edit play_record.sh to change the number of consecutive trials to record, and which directory to save them in.

Edit play_sound.sh to change which audio file to play before recording.

$ git clone https://github.com/hisroar/HomeAssistantSonar.git
$ cd HomeAssistantSonar
$ ./play_record.sh

Feature extraction and classfication

Below is a list of Matlab files and what they do. It is recommended to run them in the order provided below. Data (trial folders) need to be in the same directory as all the Matlab files.

Code is commented for ease of re-use.

Matlab file	Description
`FolderSetupCleanup.m`	Cleans up previous `FeatureExtract/` directory and creates new one. Only needs to be run once unless data is changed.
`GenerateMLData.m`	Reads `.wav` files and outputs all feature data to `.csv` files. Only needs to be run once unless data is changed.
`GenerateFeatureMatrix.m`	Reads data from `.csv` files and creates data matrix for Matlab to read. Only needs to be run once unless data is changed or different pruned data is desired.
`MLEval.m`	Reads in feature matrix and labels and runs various ML classifiers on the data. Outputs accuracies to `ML_Accuracy.txt`.
`MLCostTest.m`	Test impact of cost matrix on various classifiers.
`PCA.m`	Test principle component analysis on various classifiers.
`FeatureExtract.m`	Function used by `GenerateMLData.m` to process `.wav` files.
`FeatureSelect.m`	Function used to run Relief-F algorithm for feature matrix.

License

MIT

msrivastava / HomeAssistantSonar