NumberDetectionAndRecognitionUsingCNN

Computer vision project used to recognize and detect street house numbers in videos and images.

Purpose

This was the final project for my Computer Vision class. The goal was to use the http://ufldl.stanford.edu/housenumbers/ to

1. find the numbers in the image or video
2. Draw a box around it
3. Then recognize which digits appear in the image/video

You can find my report here. https://github.com/casanas10/NumberDetectionAndRecognitionUsingCNN/blob/master/cv_proj_report.pdf

TechStack

• Python
• Computer Vision
• OpenCV
• Deep Learning using Keras
• Convolutional Neural Network(CNN)
• Numpy
• Matplotlib

Outcome

• learned how to create deep neural networks using Keras
• implemented a sliding window approach to detecting street numbers
• created several CNNs and compared their results
• compared different plots

Running File

• first download weights for the CNN models https://drive.google.com/open?id=1-CzrTm7OJpY_WdYyc_SaltTnv5L4Bp9r

python run_project.py

If you want to input own image, go the run_project.py file and scroll to bottom to see main function. Check that the process_images is not commented and put your images in the input_images/TA_Testing_Folder. Then run the script just as before.

If you would like to run a video. Uncomment the video processing funtion. Put the video inside the input_videos directory and update name of the video in the run_project file. Run the script again.

Directories:

graded_images folder has the 5 images for the report

TA_Testing_Folder used to input the images you want to analyze

input_videos folder used for input the TA will like to test

output folder is where video will be after processing

weights folder contains all the weights of each model

Links:

Video link: https://youtu.be/4h1WYyUv6G8