Train Coach Number Detector

This project implements object detection and Optical Character Recognition (OCR) to extract train coach numbers from images using YOLO (You Only Look Once) and Tesseract.

Images

Features

• YOLO Model: Utilizes a pretrained YOLO model for object detection in images.
• OCR with Tesseract: Extracts text from the detected regions using Tesseract.
• Image Rotation: Enhances OCR accuracy by rotating images at specified angles to improve text readability.

Setup

Prerequisites

• Python 3.x
• OpenCV
• NumPy
• Matplotlib
• Pytesseract
• Flask

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/your-repo-name.git
   cd your-repo-name

2. Install dependencies:

   pip install -r requirements.txt

3. Download the YOLO model:

   • Ensure the best.onnx model is located in ./static/models/.

Running the Application

1. Start the Flask server:

   python app.py

2. Open your browser and navigate to http://127.0.0.1:5000/.

Usage

1. Upload an Image:

   • Click on "Choose File" to select an image containing train coach numbers.
   • Click on "Upload" to process the image.

2. View Results:

   • The processed image with detected regions and extracted text will be displayed on the web page.

Image Rotation for Enhanced OCR

To improve OCR accuracy, images are rotated at specified angles. This adjustment helps Tesseract recognize text more accurately by aligning the train numbers optimally.

angles = [-10, -8, -6, -4, -2, 2, 4, 6, 8, 10]

Each image undergoes rotation through these angles, and OCR is performed on each rotated image. The image yielding the highest confidence score from Tesseract is selected as the best result.

Key Functions

get_detections(img, net)

Converts the image to YOLO format and retrieves predictions from the YOLO model.

non_maximum_supression(input_image, detections)

Filters detections based on confidence and probability scores, applying Non-Maximum Suppression (NMS).

extract_text(image, bbox)

Extracts text from the specified bounding box region of the image using Tesseract.

drawings(image, boxes_np, confidences_np, index)

Draws bounding boxes and extracted text on the image.

rotate_y(img, y)

Rotates the image by the specified angle.

rotate_z(img, angle)

Rotates the image by the specified angle around the center using affine transformation.

process_images_with_yolo(img, net, angles)

Processes the image with YOLO and OCR, iterating over different rotation angles.

object_detection(path, filename)

Main function to handle image reading, processing, and saving the result.

Flask Application

Routes

• /: Main route to upload and display images.

Running the Server

To run the server, execute:

python app.py