Barcode Decode and Detect

This repository demonstrates barcode detection and decoding using both rule-based approaches with OpenCV and pyzbar, as well as deep learning methods with Mask-RCNN instance segmentation using the detectron2 framework.

Dataset

• The dataset used for training the deep learning model consists of 12-14 annotated images tailored to the project's requirements.
• Additional images are included for testing and validation.

Dataset Link

Results

• Results of the detection and decoding are available in the following Google Drive folder: Results

To explore the code and detailed explanation, switch to the project directory:

cd {Directory name}

Rule Based Approach

A rule-based approach using OpenCV and pyzbar is implemented with hard-coded kernel functions.

• Rule Based Approach Repository

Deep Learning Approach

For deep learning-based detection, Mask-RCNN instance segmentation model is employed using detectron2.

MRCNN No Barcode

• Detects objects without visible barcodes and outlines them with red bounding boxes.

• MRCNN No Barcode Repository

Bounding Box Around All Objects

• Detects all objects and draws black bounding boxes around them.

• Bounding Box Around All Object Repository

MRCNN Complete or Partial Barcode

• Draws blue bounding boxes around complete visible barcodes and yellow bounding boxes otherwise.

• MRCNN Complete or Partial Barcode Repository

Result Analysis

Given the small dataset size (10-12 images per case), the results indicate potential overfitting. To improve the model's performance, consider the following actions:

• Increase the size and diversity of the dataset.
• Apply data augmentation techniques to enhance generalization.
• Experiment with alternative deep learning models suited for barcode detection.

Pretrained models for barcode detection were considered but avoided due to dataset mismatch with the provided dataset.