Schedule Pickup of Glidewell Orders Using DeepLens


Glidewell Laboratories provide a wide range of dental products and services. Thoursands of orders are placed on a daily basis across the US and most of calls received by our customer serivce are order pickup request from dentists. We want to develp a next-generation customer management system to automatically schedule pickup of Glidewell orders.

What it does

This project automatically schedules pickup of Glidewell dental products using the AWS DeepLens device. When a dentist from a dental practice wishes to place a new order and schedule a pickup, s/he holds up a Glidewell dental product and stands in front of a DeepLens camera. After recognizing the dentist's identity and the product information, the system automatically places the order, schedules the pickup, and send out the confirmation email.

How we built it

  • Image Datasets

The balanced training dataset encompasses four different classes, i.e. BruxZir S, BruxZir M, BruxZir L, and no BruxZir, each of which contains approximately 500 images.

  • Programming Language & Tools

    • Python, MXNet
    • AWS: SageMaker, Lambda Function, DeepLens, Amazon Rekognition (Facial Recognition)
  • Product Classification

We utilizes the Amazon SageMaker image classification algorithm to train a classification model for Glidewell dental products. The last layer of a pre-trained, 34-layer convolutional neural network (ResNet) was adjusted to accommodate the new labelling rules. The input images have the same dimensions of 3 * 512 * 512. After training and optimization, the model takes the video stream from DeepLens and publishes the classification results to AWS IoT.

  • Face Detection and Recognition

The DeepLens sample project provides a face detection model to allow the AWS DeepLens device to detect the faces of people. The model takes the video stream from DeepLens as input and publishes the images of faces that it detects. Subsequently, AWS facial recognition identifies a person using the provided repository of face images.

Workflow diagram

alt text

Challenges we ran into

  • Dataflow across different applications

  • Mismatch of image dimensions between the training dataset and the DeepLens video stream

Accomplishments that we're proud of

  • A new idea about the order/customer management system

  • Faciliate the operation efficiency

What we learned

  • DeepLens and SageMaker

  • Integration of various resources/applications to build a machine learning project

What's next for Schedule Pickup of Glidewell Products Using DeepLens

  • Create a order in the customer managemnet system

  • Schedule a pickup for the package at the dental Practice based on customer preferences (e.g. time, carrier) in the database

  • Add the text notification system

  • Prepare more training images

  • Include more product categories

  • Build an object detection model for dental products

Built With

Share this project: