Smart Library Shelf Auditing System

Overview

Contributed to the development of an automated library shelf auditing system using computer vision for real-time book detection, identification, and shelf auditing. This project was developed for the National Library of Luxembourg, enabling efficient visual inventory management across their extensive collection.

Project Context

I worked on vision-based inventory management solutions, deploying YOLO and feature matching techniques to automate visual auditing workflows.

System Architecture

graph TD
    classDef input fill:#e1f5fe,stroke:#01579b,stroke-width:2px,color:#000000;
    classDef process fill:#f3e5f5,stroke:#4a148c,stroke-width:2px,color:#000000;
    classDef result fill:#e8f5e9,stroke:#1b5e20,stroke-width:2px,color:#000000;
    classDef feature fill:#fff3e0,stroke:#e65100,stroke-width:2px,color:#000000;

    A(["Camera Input<br/>(Shelf Image/Video)"]) --> B["YOLO Book Detection<br/>(Spine Localization)"]
    
    subgraph Identification ["Identification Methods"]
        direction LR
        C["Data Matrix Decode<br/>(Primary Path)"]
        D["Feature-Based<br/>Identification<br/>(Fallback Path)"]
    end
    
    B --> C
    B --> D
    
    D --> E["Multi-Descriptor Matching"]
    
    subgraph Descriptors ["Feature Extractors"]
        direction LR
        F["SIFT"] ~~~ G["LBP"] ~~~ H["Gabor"] ~~~ I["HOG"]
    end
    
    E --> Descriptors
    
    C --> J(["Inventory Database<br/>(Present / Missing / Wrong Order / Misshelved)"])
    Descriptors --> J
    
    class A input;
    class B,C,D,E process;
    class F,G,H,I feature;
    class J result;

Detection Pipeline

Primary Path: Barcode Decoding

• Data Matrix / ISBN barcode detection
• Direct database lookup for instant identification
• Works best for books with visible, undamaged barcodes

Fallback Path: Feature Matching

When barcodes are unreadable, the system uses multi-descriptor matching with books in database:

Inventory Status Detection

Technologies Used