These scripts can be used with the accompanying website OR on their own. They are structured in a tree branching out from the main manager file, which is the entry point for most operations against the application. This document, designed for developers looking to use facial or gesture recognition in their applications, details the various operations you can undertake using this library.
Other than the manager, the library consists of facial and gesture recognition modules as well as a common module that all the scripts utilise to some extent or another. Some of the scripts can be executed separate to the manager entirely (execute python <script-name>.py -h to see if it is permitted against the script) but the manager is the main entry point for all operations.
The manager.py is the main file of this library. It takes in a set of CLI arguments and returns the result in a response file, as well as the console log. As such, this library is designed to be run in a terminal window but can also be executed as a subprocess in another application. The script uses argparse to take in input arguments from the command line. To view the updated helper doc similar to the one below, execute python manager.py -h
usage: manager.py [-h] -a {create,edit,delete,compare,gesture} [-f FACE] [-l LOCK [LOCK ...]] [-u UNLOCK [UNLOCK ...]] [-n NAME]
[-t TIMEOUT] [-p PROFILE] [-m]
Welcome to the eye of horus facial and gesture recognition authentication system! Please see the command options below for the usage of this tool outside of a website environment.
optional arguments:
-h, --help show this help message and exit
-a {create,edit,delete,compare,gesture}, --action {create,edit,delete,compare,gesture}
Only one action can be performed at one time:
create: Creates a new user --profile in s3 and uploads and indexes the --face file alongside the ----lock-gestures (OPTIONAL) and --unlock-gestures image files. --name can optionally be added if the name of the --face file is not what it should be in S3.
edit: Edits a user --profile account's --face, --lock or --unlock feature. If you wish to delete your lock combination, specify --lock DELETE in lieu of entering a combination of gesture types to change your combination to. Note: It is not possible to rename a user --profile. Please delete your account and create a new one if you wish to do so.
delete: Deletes a user --profile account inside S3 by doing the reverse of --action create.
compare: Starts streaming and executes the facial comparison library against ALL users in the database. Alternatively, you can specify a --face to compare against a --profile's. Or you can specify a --profile on it's own to compare the captured face with that profile's stored face. You can alter the length of the stream search timeout with --timeout.
gesture: Takes a number of --lock OR --unlock images as input for authenticating with the gesture recognition client against the user --profile.
-f FACE, --face FACE Path to the jpg or png image file to use as your facial recognition face to compare against when running the kinesis stream
-l LOCK [LOCK ...], --lock LOCK [LOCK ...]
ABSOLUTE Paths to jpg or png image files (seperated with spaces) to use as the --profile user's lock gesture recognition combination (OPTIONAL). Use with -a edit/create to construct a new combination or to delete an existing one by specifying DELETE in lieu OR with -a gesture to attempt to authenticate with the matching gestures
-u UNLOCK [UNLOCK ...], --unlock UNLOCK [UNLOCK ...]
ABSOLUTE Paths to jpg or png image files (seperated with spaces) to use as the --profile user's unlock gesture recognition combination. Use with -a edit/create to construct a new combination OR with -a gesture to attempt to authenticate with the matching gestures
-n NAME, --name NAME S3 name of the face image to be uploaded. This is what the image will be stored as in S3. If not specified, the filename passed to --file is used instead.
-t TIMEOUT, --timeout TIMEOUT
Timeout (in seconds) for the stream to timeout after not finding a face during comparison
Used with -a compare, default is 20
-p PROFILE, --profile PROFILE
Username to perform the -a action upon. Result depends on the action chosen
-m, --maintain If this parameter is set, the gesture recognition project will not be shutdown after rekognition is complete (only applicable with -a create,gesture,edit)
Depending on the actions you wish to take (e.g. create, edit, delete compare or gesture), you need to pass certain parameters to the script in order for it to complete successfully. For example, to create a new user called foobar, you will need to pass the profile's username, lock & unlock combinations, face and create action. Like so:
Note: The paths to files must be absolute paths
python manager.py -a create -f /Users/someuser/Documents/myFace.jpg -p foobar -l /Users/someuser/Documents/my_lock_gesture_1.jpg /Users/someuser/Documents/my_lock_gesture_2.jpg /Users/someuser/Documents/my_lock_gesture_3.jpg /Users/someuser/Documents/my_lock_gesture_4.jpg /Users/someuser/Documents/my_lock_gesture_5.jpg -u /Users/someuser/Documents/my_unlock_gesture_1.jpg /Users/someuser/Documents/my_unlock_gesture_2.jpg /Users/someuser/Documents/my_unlock_gesture_3.jpg /Users/someuser/Documents/my_unlock_gesture_4.jpg /Users/someuser/Documents/my_unlock_gesture_5.jpg
Some actions are optional and provide helpful configurable options for the user. For example, the -t option will extend the timeout of the Kinesis facial recognition in case slow or unstable connections are expected.
The scripts use the AWS boto3 SDK for Python that allows them to communicate with the AWS services that conduct the recognition and store the user data. The main job of the library is to parse and operate on data returned from boto3 that may not reach the frontend of the applications, being in a sort of helper position usually filled by a nodejs or php server.
Once the scripts have finished carrying out their tasks, they will produce a response.json file in this directory with the details of how the script performed. This makes it easier for separate applications not running a Python engine to reliably parse the response of the script.
The structure will always be the same, no matter the operation or amount of debug log the script produced. For example, running a face comparison with a source image and the stored image for the user morgan, produces this response:
{
"TYPE": "SUCCESS | ERROR (the result of the script in a single word)",
"MESSAGE": "Input face /Users/morgan/Downloads/morgan.jpg matched successfully with stored user's morgan face (a short explanation detailing what the result was)",
"CONTENT": "null (null, if it's not needed, or a longer explanation or large object that would be useful in debugging what the problem is or investigating further the object of a success)",
"CODE": 0
}The response file will also provide an exit CODE value that the script escaped with in integer format. Anything other than 0 is considered an unsuccessful exit. See Codes for what each code stands for.
No matter the script, all will exit with one of the following codes. For more information on any errors, check the MESSAGE and CONTENT fields of the response file.
- Successful exit
- General Error
- Something Wasn't Found
- Something Failed To Upload
- Something Failed To Delete
- Kinesis stream Failed To Start
- Kinesis stream Failed to Die (how dare it)
- File Is Of An Incompatible Type
- File Does Not Exist Locally
- File Does Not Exist In S3
- Stream Timeout Was Fired or face found did not match
- Stream Wasn't Found
- Stream Was Interrupted
- Invalid Action Type
- Custom Label Project Failed To Start
- Custom Label Project Failed To Start Within The Timeout
- Custom Label Project Failed To Die (How incredibly rude that it's not allowed to live)
- No recognised gesture was detected
- Gesture was detected but it was not in the user's combination and/or in in the right position of the user's combination
- Stream Timeout Failed To Fire
- Rule Violation: All gestures of the specific locktype are the same
- Rule Violation: The locking gesture is the same as the unlocking gesture
- Rule Violation: One gesture combination is the same as the other when reversed
- Custom Label Project is shutting down or starting up
- Given image is too large for detection of custom labels
- ClientError on image processing of custom labels. Likelihood is too large to even send with detect_custom_labels
- User gesture combination api is rate-limited
- Captured face in stream does not match the user's face
- Rule Violation: Given gesture combination for the specific locktype is too short (minimum combination length = 4)