At SkyWest Airlines, every two weeks we have a new class of pilots, and a new class of Flight Attendants. These classes can have an excess of 60 employees each time. The process of taking pictures and trying to recognize the employee to add their photo to our server and get ID Badges printed is currently taking much longer than it should, and frankly that’s where my brain starts churning with ideas to create a better way.
This experience in still a concept and has yet to receive the development power to make it a reality, but uses Artificial Intelligence and Google’s Vision API to function.
First the user uploads the batch of employee images. Taken with a name sticker that includes the employee number in the sticker.
Once the images have been updloaded Google Vision’s “Identify Text” will scan the image to find the text on their name badges. It will return separate variables with the name and other information found on the sticker, and the variables will be validated to only store the 5 digit number which has a # before it. All other data can be dumped.
Any photos where a number couldn’t be pulled automatically will be confirmed by the user later. The image location of the numbers identified will be saved to crop the image for user confirmation.
Once the photos have all been processed they will then be confirmed by the user and any images that couldn’t be recognized will be highlighted in red for the user to manually identify. If the user can’t tell by the image alone, a click on the image will bring up a lightbox that supports scroll zooming with the full image so the face can be identified.
Duplicate images will be weeded out in the next stage if the employee numbers are visible. The edit button will allow for manual deletion of a duplicate image if needed before the next step.
At this stage if all the numbers have been identified and there are duplicate numbers the user can manually select the best image.
When all numbers have been confirmed and all duplicates resolved the user will get a summary of the number of ID photos to be matched to the class roster before submission.
Images will then be cropped to the employee’s face using “Face Detect” from the Google Vision API. Once that completes the photos will be submitted to the queue for badges to be printed and to our employee database. From there the user can either begin a download a compressed copy of all the completed images, or start a new batch.
This is all conceptual at the moment and a few issues with validation and possible facial recognition have yet to be resolved.
In the meantime you can click through a prototype to see how the tool would work.