Intelligent Parcel Sorting System Using DIGIPIN-Based Edge AI and OCR

Abstract

The rapid expansion of e-commerce has created unprecedented demand for high-throughput, error-free parcel sorting in logistics hubs. Conventional sorting systems that rely on barcodes or QR codes introduce a redundant encoding step and fail when labels are damaged or misaligned. This paper presents a fully automated, cloud-independent parcel sorting conveyor system that leverages Edge Artificial Intelligence and pure Optical Character Recognition (OCR) to read a 10-character alphanumeric DIGIPIN (Digital Postal Index Number) directly from package surfaces. The system employs a dual-node distributed architecture: a Raspberry Pi 5 acting as the cognitive Edge AI node for image processing, OCR inference, and GUI rendering, and an Arduino UNO serving as the deterministic real-time execution node for sensor polling, motor control, and servo actuation. Node communication is achieved wirelessly via an HC-05 Bluetooth serial link. Prototype testing demonstrated 100% sorting accuracy on clean labels and 86.7% accuracy across real-world varied label conditions, achieving a cycle time of approximately 4 seconds (?900 parcels/hour). The proposed system provides a scalable, low-cost alternative to cloud-dependent industrial sorters suitable for India\'s expanding postal and logistics networks.

Introduction

The rapid growth of e-commerce has significantly increased the demand for fast and accurate package sorting in logistics and supply chain operations. Traditional sorting systems relying on manual labor, barcodes, and QR codes face limitations such as slow processing, human errors, label damage, and alignment issues. To address these challenges, this project proposes a Smart Package Sorting System that combines Edge AI, computer vision, and the DIGIPIN (Digital Postal Index Number) framework for automated parcel routing.

Problem Statement

Existing package sorting methods suffer from:

• Manual sorting: Time-consuming, labor-intensive, and prone to errors.
• Barcode systems: Require precise alignment and fail when labels are damaged.
• QR-code systems: Sensitive to smudging, wrapping, and physical wear.
• Dependence on intermediary machine-readable markers rather than directly reading printed text.

Proposed Solution

The Smart Package Sorting System uses Optical Character Recognition (OCR) to directly read the 10-character alphanumeric DIGIPIN printed on packages. This eliminates the need for barcodes or QR codes and enables intelligent package routing based solely on human-readable text.

System Architecture

The system employs a dual-node distributed architecture:

1. Cognitive Node – Raspberry Pi 5
   • Captures package images using a webcam.
   • Performs OCR and routing decisions locally using Edge AI.
   • Hosts a graphical user interface (GUI).
2. Execution Node – Arduino UNO
   • Controls conveyor operation.
   • Monitors package detection through an IR sensor.
   • Operates servo motors for package sorting.
3. Communication Layer
   • Uses an HC-05 Bluetooth module for wireless communication between Raspberry Pi and Arduino.

Hardware Components

Key components include:

• Raspberry Pi 5
• HD Webcam
• Arduino UNO
• HC-05 Bluetooth Module
• IR Proximity Sensor
• DC Gear Motor
• MG90S Servo Motors
• Conveyor Belt System

Working Principle

The sorting process follows these steps:

1. Conveyor runs continuously.
2. IR sensor detects an incoming package.
3. Conveyor stops beneath the camera.
4. Raspberry Pi captures an image.
5. OCR extracts the DIGIPIN from the package.
6. The system identifies the destination zone from the DIGIPIN.
7. A routing command is sent to the Arduino via Bluetooth.
8. The appropriate servo arm diverts the package into the correct bin.
9. The system resets and waits for the next package.

For demonstration, packages are sorted into:

• Peelamedu
• Mettupalayam

OCR and AI Pipeline

The software uses a hybrid two-stage approach:

• Primary Stage: QR-code decoding for rapid processing when available.
• Fallback Stage: AI-powered OCR extracts the DIGIPIN from captured images when QR codes are absent or unreadable.

The OCR process includes:

• Image capture
• Grayscale conversion
• Thresholding and preprocessing
• Text extraction
• DIGIPIN identification using pattern matching

Literature Survey Insights

Previous research demonstrated:

• Machine-vision parcel sorting using OCR.
• Edge AI for low-latency logistics automation.
• OCR-based postal systems.
• Distributed embedded control architectures.
• IoT-enabled conveyor sorting systems.
• Bluetooth communication for industrial automation.
• AI-driven robotic sorting solutions.

However, most existing systems rely on centralized processing, barcodes, or expensive robotic infrastructure. The proposed system improves upon these by offering:

• Edge-based real-time processing.
• Marker-free package identification.
• Low-cost hardware implementation.
• Faster and more flexible deployment.

Key Advantages

• Eliminates dependency on barcodes and QR codes.
• Real-time OCR-based package identification.
• Reduced sorting errors and labor requirements.
• Low-latency Edge AI processing.
• Wireless and scalable architecture.
• Cost-effective solution for small and medium logistics centers.
• Improved reliability when labels are damaged or poorly aligned.

Conclusion

This paper presented a low-cost, cloud automated parcel sorting system that routes packages based on the DIGIPIN standard using Edge AI and a pure OCR pipeline. The Master-Slave Cyber-Physical architecture effectively segregates computational and real-time mechanical tasks, preventing processing bottlenecks. Built entirely from off-the-shelf components at a total cost of ?24,560, the prototype demonstrated 100% accuracy on clean labels and 86.7% under varied real-world conditions, with an average throughput of approximately 900 parcels/hour. The system proves that Edge AI-driven DIGIPIN sorting is technically feasible, economically viable, and ready for adaptation within India\'s evolving logistics networks. Future work will focus on: (1) extending routing coverage to additional DIGIPIN zones by updating the software dictionary; (2) integrating transformer-based OCR models for handwritten label support; (3) adding a Hailo-8 NPU via PCIe to reduce OCR processing time from ~3 s to under 0.5 s; and (4) scaling horizontally to multiple parallel conveyor belts reporting to a central local network hub.

References

[1] M. Chen and R. Wang, \"Edge AI-Based Real-Time Object Detection for Logistics Automation,\" IEEE Access, vol. 10, pp. 112034-112045, 2022. [2] A. Kumar and P. Singh, \"Smart Conveyor Systems Using IoT and Automation,\" Int. J. Engineering Research, vol. 8, no. 4, pp. 211-217, 2019. [3] S. Ramesh and K. Priya, \"Optical Character Recognition for Automated Postal Systems,\" Int. J. Computer Applications, vol. 183, no. 12, pp. 20-26, 2021. [4] J. Smith and D. Brown, \"Distributed Embedded Systems for Industrial Automation,\" IEEE Transactions on Industrial Electronics, vol. 67, no. 9, pp. 7843-7852, 2020. [5] H. Tanaka and Y. Nakamura, \"Autonomous Sorting Systems Using AI and Robotics,\" IEEE Robotics and Automation Letters, vol. 3, no. 3, pp. 1841-1848, 2018. [6] Z. Wei and L. Yong, \"Intelligent Parcel Sorting System Using Machine Vision,\" Journal of Logistics Automation, vol. 8, no. 2, pp. 45-52, 2023.

Copyright

Copyright © 2026 Hemalatha ., Srinikesh , Prajit , Venkat Srinath, Surendran . This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.