Automatically Detecting Damage on Shipping Pallets Aashir Tuladhar , Nicholas Hubbard , and Andre Mossi 1
1
Problem Damaged pallets are a costly issue that can be unavoidable in warehouse operations. Manual inspection is inefficient and laborintensive, often leading to damaged good being accepted. Solution We’ve developed an automated pallet damage detection system that uses computer vision and machine learning to inspect pallets in real time as they enter through a dock door. The system captures live video, runs AI-based analysis, and flags damage as the pallet enters a facility. It then records these results into a tracking system, which can be accessed by warehouse managers. Key Features •
•
•
•
•
1
1
Computer and Information Science
Impact •
•
•
•
•
•
•
AI Detection: As more samples of pallets arrive, the AI model gets better over time. •
Dashboard Interface: Allows warehouse managers to view historical data, detection logs, and trends over time. Continuous Improvement: Our solution allows the AI component of our system to continuously get better with every new pallet we detect.
Improved safety and product integrity Better supplier accountability with visual evidence Reduced liability
Technology
Real-Time Detection: Provides instant alerts for damaged pallets during intake.
Seamless Integration: Designed to work with existing loading dock workflows.
Reduced labor costs and inspection time
•
Our model takes in thousands of sample images, which we label with boxes using Label Studio, and then train a new model from it using the Ultralytics YOLO platform. A Multi-Camera vision system, powered by four Seeed Studio reCameras provide live video streaming and a trigger service that starts the damage detection process. The reCamera video is streamed and stored locally on a Seeed Studio reServer Industrial (a customized NVIDIA Jetson Orin NX), where an NVIDIA DeepStream pipeline detects pallet damage in real time. Detected frames are then uploaded to Azure Blob Storage, and attributes of each frame (like timestamp, detections, camera source, etc.) are stored in a TimescaleDB database. A custom React-powered dashboard gives warehouse managers a searchable, time-stamped record of pallet inspections with images and metadata.