EyeTracking is a Swift Package that makes it easy to use ARKit's eye and facial tracking data, designed for use in Educational Technology research.
- Create and end
Sessions at any time - 60 fps stream of gaze location, or where the user is looking at the screen
- Configure any number of
ARFaceAnchor.BlendShapeLocations to record - Data persisted in SQLite with GRDB.swift
- Data import/export in JSON
DataorStringobjects - Live view of gaze location with a customizable
UIView - Fixation point and scan path visualizations
- iOS 13.2+
- Xcode 11.0+
- Swift 5.0+
- iPhone X or older / iPad Pro 2018 or older Note: On release of iOS 14 and ARKit 4, all devices with an A12 Bionic or later will be supported.
Install EyeTracking using Swift Package Manager either through Xcode 11+ directly or as a dependency of your own Swift Package. To do the latter, add EyeTracking to the dependencies value in your Package.swift file:
dependencies: [
.package(url: "https://github.com/kyle-fox/ios-eye-tracking", .upToNextMajor(from: "1.0.0"))
]This is the only officially supported installation method - if you wish to use the code directly, you are free to manually integrate EyeTracking into your project.
ARKit's facial tracking system relies on the front facing TrueDepth camera to capture its data, so you must provide a permission request in your app's Info.plist's NSCameraUsageDescription. This will be presented to the user the first time you begin an EyeTracking session, and they will be able to deny this permission. See the example application for an example of such request, and when using for research, be sure to inform participants that it is necessary to accept this permission request, when presented.
EyeTracking can be used anywhere within your app, and it is recommended that you configure and store a reference to the EyeTracking class where you wish to use it. To do this, you may simply initialize the class by providing a Configuration:
let eyeTracking = EyeTracking(configuration: Configuration(appID: "ios-eye-tracking-example"))You must provide an appID upon creation of the class - this is stored in each Session as a means of identifying the app from which the data was collected. Optionally, you may configure any number of ARFaceAnchor.BlendShapeLocations and EyeTracking will record that data in a Session alongside the gaze location data. This example will track the blink data for both eyes (see Apple's documentation for more options and information on the data):
let eyeTracking = EyeTracking(configuration: Configuration(appID: "ios-eye-tracking-example", blendShapes: [.eyeBlinkLeft, .eyeBlinkRight]))Now that you have the class configured, you are free to start recording data at any time by calling:
eyeTracking.startSession()This will begin a session and start streaming all data from ARKit into a Session object. Each Session will have a UUID, the appID provided at initialization, a UNIX timestamp for the beginning of the session, and device information, including model, screen size, OS name, and OS version. By default, the data streams at 60 fps, so be sure to test your app for memory use, as the size of this data can grow quickly, depending on how many ARFaceAnchor.BlendShapeLocations are configured.
Note: This package only supports running 1 Session at a time.
When you would like to end the Session that is currently running, you may do so at any time by calling:
eyeTracking.endSession()This will stop the current Session and write all its data to the database, which you may then export at any time.
EyeTracking supports exporting its data in JSON format, in either Data or String types. You may export either a single session, using it's UUID, or all sessions in the database. For all export functions, you may optionally provide a JSONEncoder.KeyEncodingStrategy to format the data's keys. The default functionality is to use the original, camelCase keys. All of these functions are marked throws, passing on their values from their underlying Codable functions. Examples:
// Exports this session as `Data`, using the default Keys
let dataSession = try? eyeTracking.export(sessionID: "8136AD7E-7262-4F07-A554-2605506B985D")
// Exports this session as a `String`, converting the keys to snake case
let stringSession = try? eyeTracking.exportString(sessionID: "8136AD7E-7262-4F07-A554-2605506B985D", with: .convertToSnakeCase)
// Exports all `Session`s as `Data`, using the default keys
let dataSessions = try? eyeTracking.exportAll()
// Exports all `Session`s as a `String`, converting the keys to snake case
let stringSessions = try? eyeTracking.exportAllString(with: .convertToSnakeCase)Importing data may be useful for viewing Session data that was collected on another device. These functions are similar to exporting, and the data you provide will be written to the SQLite database for use in the package's visualizations. Sessions may be in either Data or String types, but they must be in JSON format - it is safest to import data that has been exported with this package. Similar to exporting, you may provide an optional JSONDecoder.KeyDecodingStrategy if you have keys in snake case or others. All of these functions are marked throws, passing on their values from their underlying Codable functions. Examples:
// Imports a `Session` from a JSON `Data` object
try? eyeTracking.importSession(from: data)
// Imports a `Session` from a JSON `String`, converting the data's keys from snake case.
try? eyeTracking.importSession(from: jsonString, with: .convertFromSnakeCase)
// Imports an array of `Session`s from a JSON `Data` object, converting the data's keys from snake case.
try? eyeTracking.importSessions(from: data, with: .convertFromSnakeCase)
// Imports an array of `Session`s from a JSON `String`
try? eyeTracking.importSessions(from: jsonString)Deleting data is straightforward and offers a few options:
// Deletes a given `Session` object from the database.
try? eyeTracking.delete(session)
// Delete all `Session` objects in the database.
try? eyeTracking.deleteAll()
// Delete the entire database and everything in it.
try? eyeTracking.deleteDatabase()To get an idea of the data structure, below is a truncated example of the JSON data output. For an example of full data output, see example_json_output.json.
{
"beginTime": 1595626843.445204,
"id": "3BE54760-F1E8-4799-A3CB-78806270E72A",
"appID": "ios-eye-tracking-example",
"endTime": 1595626853.5600219,
"scanPath": [
{
"y": 297.72345417737961,
"timestamp": 1595626844.0514789,
"x": 479.91125732660294,
"orientation": 3
},
{
"y": 287.79990407824516,
"timestamp": 1595626844.068157,
"x": 484.62997680902481,
"orientation": 3
},
...
],
"blendShapes": {
"eyeBlink_L": [
{
"blendShapeLocation": "eyeBlink_L",
"value": 0,
"timestamp": 1595626844.0514789,
"orientation": 3
},
{
"blendShapeLocation": "eyeBlink_L",
"value": 0,
"timestamp": 1595626844.068157,
"orientation": 3
},
...
],
"eyeBlink_R": [
{
"blendShapeLocation": "eyeBlink_R",
"value": 0,
"timestamp": 1595626844.0514789,
"orientation": 3
},
{
"blendShapeLocation": "eyeBlink_R",
"value": 0,
"timestamp": 1595626844.068157,
"orientation": 3
},
...
]
},
"deviceInfo": {
"systemVersion": "13.5.1",
"systemName": "iOS",
"model": "iPad8,1",
"screenSize": [834, 1194]
}
}All logging is done through Apple's lightweight os.log system, and logs can be viewed in either the Xcode console or macOS's Console.app. You do not need to be connected to a debug session to view these logs - you may use Console.app to view the logs of any device on the network that is using EyeTracking. By default, only critical issues are logged, but you may optionally enable debug logging for all collected data by setting:
eyeTracking.loggingEnabled = trueFor debugging purposed, you may wish to view the live gaze location on screen. By default, this uses a blue, 30px by 30px UIView called pointer, but you may modify this view's configuration at any time before display. To do so, you may call:
// Optionally make any changes to the pointer
eyeTracking.pointer.backgroundColor = .red
// Begin displaying the pointer
eyeTracking.showPointer()Similarly, you may remove the display at any time:
eyeTracking.hidePointer()EyeTracking supports the ability to playback the scanpath, or the path the user's eyes followed during a Session. You may set the color of the path that is drawn onscreen and choose whether or not to animate the path. By default, the duration of its animation is the duration of the Session itself, so this will serve as a playback of the Session as it happened, but you may optionally provide your own duration. You may display this visualization at any time, given you have a Session's unique identifier, like so:
EyeTracking.displayScanpath(for: "8136AD7E-7262-4F07-A554-2605506B985D", animated: true)