Regional-Language-Detector

A regional language and location detecting application using a tri-combination of Google's TensorFlow Object Detection API, Convolutional Recurrent Neural Network (CRNN) and a 2-layered Neural Network for language classification.

The Application uses three modules in sequential manner:

1. Google's TensorFlow Object Detection API for Text localisation
2. CRNN for text Recognition
3. 2-layered Neural Network for language classification Location is determined by predicted text using Geotext

Starting Application

• Edit the Image path in main.py

#Image-Location-Detector/
python3 main.py

Requirements

• [Python 3.6]
• [TensorFlow 1.2]
• [PyTorch]
• [OpenCv] - Either build from source or

sudo apt-get install python-opencv

• [Geotext]

pip3 install geotext

Downloads

wget https://www.dropbox.com/s/l0vo83hmvv2aipn/crnn.pth

Notes

• Currently ssd_mobilenet_v1_coco_11_06_2017 model is being used - boundry box prediction for text detection will be poor
• Try to build your own custom dataset using labelImg

Todos

• Train custom dataset for Text detection
• Notes on generating custom dataset

Refrences

• https://github.com/AKSHAYUBHAT/DeepVideoAnalytics/tree/master/notebooks/OCR
• https://github.com/datitran/object_detector_app