SCAN – LESS PALLETSORTING

SCAN – LESS PALLET SORTING

The customer wants the AI engine to take the images of the pallets during various stages of the inbound, put-away and outbound processes, and detect the labels of interest, decode the values of the labels (bar and QR codes) and provides it to the inventory management system.

Creating the dataset from scratch, instructed and trained the client team on-ground staff and SPOC to take videos and images of the pallets :

Warehouse with a variety of image capture devices different types of pallets and different labels used for pallets different stages of the inbound, put-away and outbound process Created a custom detection and classification model for the client using open source object detection models employing

Transfer learning and Retraining, to detect a pallet, labels of interest seen on the pallet, and identify the label and classify the label type

Extraction of the labels of interest from the image and sent it to the barcode and QR code decoder

Get the decoded values of the bar and QR codes and sent it to the Warehouse inventory management system

When storing the pallets in the warehouse, the rack level bar codes also need to be scanned and the decoded values need to be sent to the Warehouse inventory management system, during the storage process

CHALLENGES

Understand the different processes and identify the exact time to capture images

Pallets will be covered with plastic wraps since they are stored in cold storage warehouses

The storage spaces will be multi level deep, and will have low lighting conditions

TECHNOLOGIES / TOOLS

Keras, TensorFlow , OpenCV

ResNet50 used as feature pyramid network

PyZBar

MODEL HIGHLIGHTS

When given a same set of images (10 images) to Azure, Google and our algorithm, we could detect more faces than Azure and Google.

Model size wise we are 110MB and performance wise faster than Face++ and other models like MobileNet and VGG

Our model could differentiate between Twins

We use 512 feature embedding which means a lot of details about a face is captured and stored which helps in very effective recognition