miworking / xfactor_2_pickup

App Name: Pickup

What does it do:

Once the app is opened, it will go to the background immediately, so there is no UI for this app. However it will work in the background as a service, monitoring the user's gesture to decide whether the phone is opened by the owner or not.

How to kill it:

Since this app has no UI, and only works as a background service, it can only be killed manually in the settings of the system.

How it works

When the screen is turned off, all sensor data will be collected by this service automatically, and these data will be written into a local csv file in external folder like this: /storage/emulated/legacy/Android/data/edu.cmu.ebiz.pickup/files/

Once the screen is turned on, it will read these data from local files, and abstract features from them, and decide whether it is the owner that picked up this phone.

Machine learning algorithm

Decison is made basing on a random forest model which was trained before, this model file is located in assets folder. And we are using Random Forest algorithm prived by Weka, to use Weka in this app, a weka.jar is imported, and this line needs to be added to dependencies in build.gradle

compile files('libs/weka.jar')

Data format and related features

Only accelerometer and magnetic sensor data will be used, so there will be two files in this folder: 2015_08_21_23_47_55_acc.csv and 2015_08_21_23_47_55_magetic.csv ([Why don't use other sensor data?] (https://www.dropbox.com/s/bnvwc62nh7kt24q/Pickup.pptx?dl=0) [Page 2])

104 featureswill be abstracted [{Accelerometer, Magnetic} * {X,Y,Z,Magnitude} * {Mean, Std, Min, Max, Percentile25,50,75, FTT[0,1,2]}]

Related Classes

All data related classes will be placed in pattern package

• [DoubleFeatureUnit] (https://github.com/miworking/XFactor_PickupRecognition/blob/master/app/src/main/java/edu/cmu/ebiz/pickup/pattern/DoubleFeatureUnit.java) is the basic class,

• XYZFeature has 4 DoubleFeatureUnit members: X,Y,Z,V (Magnitude of X,Y,Z)

• Feature class contains 2 XYZFeature members: accelerometer and magnetic

• OnePersonData has a list of Feature as its member, and calculate features from it

What's the result?

Once the result is decided from this Random Forest model, it will be stored in isOwner as a boolean variable, and this result will be [sent out as a broadcast to "org.twinone.locker.pickup.result" Action] (https://github.com/miworking/XFactor_PickupRecognition/blob/master/app/src/main/java/edu/cmu/ebiz/pickup/AEScreenOnOffService.java#L271-L286)

and the XFactor app will register a receiver for this broadcast, and calculate a pickup_score basing on this result.

Collect data

This app can also be used to collect data for later modeling. The only difference is to comment the following line so that data will not be cleared everyone when use turns off the screen. Once data is collected, you can use TrainModel to preprocess them and generate features in a arff file, so as to be used in Weka.

After you have collected enough data, you'd better copy this folder to your desktop and rename it so that it wont be erased in the future. /storage/emulated/legacy/Android/data/edu.cmu.ebiz.pickup/files/

Here we collected data from 6 people, and put them like this

you can refer to the README of TrainModel to go further on how to use these data to traing a model