The customer is in the Cold storage and Warehousing business and would like to resolve the pain points during handling of the inbound pallets. They want the AI engine to take the images of the inbound pallets from the conveyor and present the type value and the box count to their WMS system, when the pallet reaches the profile check station.
PALLET SCANNER
-
Created the dataset from scratch, trained the client team on-ground staff and SPOC to take videos and images of the pallets.
- In various warehouses
- With a variety of image capture devices
- Different types of pallets
- Different stages of the inbound 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 and identify the various part of the pallet and classify the type
- A box detection model analyses a part of the pallet images and detects the boxes in both the side and top images
- A box counting algorithm takes the number of boxes detected on the side and top images and computes the number of boxes
CHALLENGES
- Pallets will be covered with plastic wraps since they are stored in cold storage warehouses which causes a lot of glare
TECHNOLOGIES / TOOLS
- Keras, TensorFlow
- ResNet50 used as feature pyramid network
MODEL HIGHLIGHTS
Detector and classification model for the parts of the pallet
- Full pallet
- Box portion
- Pallet base
- Type of the pallet based on the classification of the base of pallet portion