Barcode Scanning Video Dataset

Home » Case Study » Barcode Scanning Video Dataset

Project Overview:

Objective

GTS’s mission was to create a specialized video dataset designed to revolutionize barcode scanning technology. Therefore, this Barcode Scanning Video Dataset aims to enhance machine learning models, making barcode detection more efficient and accurate in various real-world scenarios. By using this dataset, developers can significantly improve the performance of their barcode scanning applications.

Scope

We carefully collected and labeled a diverse set of videos, showcasing barcodes in different settings. Consequently, this variety ensures our dataset is adaptable and boosts the effectiveness of training machine learning models. Moreover, the inclusion of various scenarios enhances the robustness of the models. Therefore, the dataset is more likely to perform well across different applications.

Sources

Retail store environments: products on shelves, cashier checkout scenarios.
Warehouses: inventory checks, package labelling.
Home settings: scanning items for online shopping apps, personal inventory.
Outdoor scenarios: scanning tickets at events, QR codes on ads.

Data Collection Metrics

Total Video Clips: 50,000
Retail Store Clips: 20,000
Warehouse Clips: 10,000
Home Setting Clips: 12,000
Outdoor Scenarios Clips: 8,000

Annotation Process

Stages

Bounding Boxes: First, we draw boxes around barcodes in each video frame. This helps to identify the exact location of each barcode.
Barcode Type Classification: Next, we label the barcode types, such as UPC, QR, and Code128. This classification is crucial for further processing.
Transcription: Additionally, we provide the exact digital equivalent of the barcode where possible. This ensures accurate data extraction.
Environmental Tags: We also mark environmental factors, such as lighting conditions, noting whether it is low-light or has glare, and any obstructions. This helps to understand the context in which the barcode is scanned.
Orientation Tags: Lastly, we note the barcode orientations, such as if the barcode is tilted or upside-down. This information is important for improving scanning accuracy.

Annotation Metrics

Total Annotations: 1,250,000 (considering average 25 frames annotated per video)
Bounding Boxes: 800,000
Barcode Classifications: 200,000
Transcriptions: 100,000
Environmental Tags: 100,000
Orientation Tags: 50,000

Quality Assurance

Stages

Expert Review: We engage experts in barcode technology to review annotations, which helps us keep high standards.
Consistency Checks: Additionally, we use automated systems to check the accuracy of transcriptions and ensure bounding boxes are correctly aligned.
Inter-annotator Agreement: Moreover, to ensure consistency, we assign overlapping sections of the dataset to multiple annotators.

QA Metrics

Annotations Reviewed by Experts: 125,000 (10% of total annotations)
Inconsistencies Identified and Rectified: 25,000 (2% of total annotations)

Conclusion

The Barcode Scanning Video Dataset project successfully brought together and annotated a diverse set of videos designed for barcode recognition. By focusing on systematic processes and quality, this dataset is set to significantly advance barcode scanning technology. Consequently, it will enable faster, more versatile, and accurate recognition in various applications.